--- schema_version: 1 rubric_version: "1.1" kind: pair canonical_url: "https://onaihq.com/compare/agents/dust-vs-librechat/" data_url: "https://onaihq.com/compare/agents/dust-vs-librechat/data.json" generated_at: 2026-10-27 last_checked_at: 2026-10-26 last_cross_checked_at: 2026-10-26 next_check_at: 2026-12-20 publisher: "onAI" --- # Dust vs LibreChat > Published by onAI. onAI also sells one of the products compared here. Findings about other products link to public sources so you can check them; statements about onAI are marked as stated by onAI. This comparison has 44 criteria with sourced findings for both products, including 44 that a second agent cross-checked for both. 51 criteria have a sourced finding for at least one product. "Sourced" findings passed publication validation: they are scoped, fresh and cite public resolved sources. "Agent cross-checked" findings also received a separate agent source pass, not necessarily a hands-on test. "Stated by onAI" marks the publisher's own statements about its product (evidence type vendor-statement), which cite no third-party source. Humans approve the rubric, templates and generator; they do not review every finding. It does not declare a winner. A missing sourced finding is not evidence that a capability is absent. - Dust source checks: 2026-10-21. - LibreChat source checks: 2026-10-21 to 2026-10-26. - Canonical HTML: https://onaihq.com/compare/agents/dust-vs-librechat/ - Machine-readable JSON: https://onaihq.com/compare/agents/dust-vs-librechat/data.json - Methodology: https://onaihq.com/compare/agents/methodology/ ## Models ### MOD-01: Model catalogue and providers Judgment rule: List each available model by provider and plan, including reasoning, fast and multimodal variants. Record whether the user picks the model, the product routes automatically, or both. A provider counts only if at least one of its current models is selectable on the scoped plan. #### Dust Verification status: Agent cross-checked (agent-cross-checked) Support: partial Dust offers per-message model choice and ordered auto-model lists, with access-tier and region controls. Availability: generally-available Delivery: native Scope: surfaces=Dust web workspace, Dust EU web workspace; version=Public source pass 2026-10-21; no logged-in deployment observed; plan=Business credit plan (Free/Pro/Max); Enterprise model access inherits the published catalogue; model-tier and EU eligibility separately constrained; add_ons=none stated; deployment_mode=multi-tenant-saas; region=US/global or EU workspace as named; EU-hosted-model restriction must be enabled separately; os=not specified. Limitations: - Documented findings only; unestablished atomic questions remain explicit gaps. - The catalogue finding is restricted to named documentation entries; conflicting Gemini4 Argon EU eligibility is an explicit gap. - providers — Which model providers are available? [supported]: Docs list OpenAI, Anthropic, Google, Mistral and Fireworks-served models; exact enabled models depend on workspace and tier. Sources: dust:dust:models, dust:dust:eu-models. - current_models — Which current models are selectable on this plan? [partial]: The model-selection document lists GPT6.1 Sol, Claude Opus5.5/Sonnet5.5, Gemini3.8 Flash and Mistral models. Gemini4 Argon is excluded from this current-EU availability finding: its release note conflicts with the EU table, and the documented authoritative picker flag was not observed. No complete current tenant catalogue is claimed. Sources: dust:dust:models, dust:dust:changelog, dust:dust:eu-models, dust:dust:pricing. - user_choice — Can users pick the model per chat? [supported]: The composer chooses a model for the next message; Agent Builder sets a persistent agent default. Sources: dust:dust:models. - auto_routing — Does the product route between models automatically, and is the chosen model shown? [supported]: Basic/Standard/Premium use ordered candidate lists and availability fallback, not prompt-dependent task routing. A replacement indicator names the fallback during a detected outage. Sources: dust:dust:models. - multimodal_input — Which models accept images, audio or files? [partial]: Conversation files support images, documents and audio transcripts; the exact input capability for every current model is not fully tabulated. Sources: dust:dust:files. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: dust:dust:models, dust:dust:eu-models, dust:dust:changelog, dust:dust:pricing, dust:dust:files #### LibreChat Verification status: Agent cross-checked (agent-cross-checked) Support: partial The model catalog is configured by the customer across first-party provider integrations and custom endpoints. Availability: generally-available Delivery: native Scope: surfaces=Customer-operated web application and server; version=Public documentation retrieved 2026-10-21; inspected source v0.8.8 / e8f3be08623663d4ad7f7241e693c94469b63bb0; plan=MIT customer-operated LibreChat; optional separately deployed services identified per answer; add_ons=none stated; deployment_mode=on-premise; region=Customer-selected infrastructure; model, OCR, search, observability and support services separately scoped; os=not specified. Limitations: - providers: Documented routes include OpenAI, Anthropic, Google, Azure OpenAI, AWS Bedrock and custom compatible services; selection requires enabled credentials and model access. - current_models: v0.8.8 release notes mention GPT-6/GPT-6.1 Sol, Claude Opus/Sonnet 5.5 and Gemini 3.8 Flash support. Actual selectable IDs still depend on the installed configuration and provider account. - auto_routing: Does the product route between models automatically, and is the chosen model shown? No automatic routing policy showing a chosen model was established, and no configured endpoint was exercised. - multimodal_input: Provider-native files support images/PDFs and some providers accept additional Office/audio/video types. Exact acceptance depends on the selected model and transport; no matrix was tested. - providers — Which model providers are available? [partial]: Documented routes include OpenAI, Anthropic, Google, Azure OpenAI, AWS Bedrock and custom compatible services; selection requires enabled credentials and model access. Sources: librechat:librechat:custom, librechat:librechat:azure, librechat:librechat:bedrock, librechat:librechat:vertex. - current_models — Which current models are selectable on this plan? [partial]: v0.8.8 release notes mention GPT-6/GPT-6.1 Sol, Claude Opus/Sonnet 5.5 and Gemini 3.8 Flash support. Actual selectable IDs still depend on the installed configuration and provider account. Sources: librechat:librechat:releases. - user_choice — Can users pick the model per chat? [supported]: Users choose an endpoint/model unless admins enforce model specs or a fixed Agent. Sources: librechat:librechat:modelspec, librechat:librechat:agents. - auto_routing — Does the product route between models automatically, and is the chosen model shown? [unknown]: auto_routing: Does the product route between models automatically, and is the chosen model shown? No automatic routing policy showing a chosen model was established, and no configured endpoint was exercised. - multimodal_input — Which models accept images, audio or files? [partial]: Provider-native files support images/PDFs and some providers accept additional Office/audio/video types. Exact acceptance depends on the selected model and transport; no matrix was tested. Sources: librechat:librechat:ocr. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: librechat:librechat:custom, librechat:librechat:azure, librechat:librechat:bedrock, librechat:librechat:vertex, librechat:librechat:releases, librechat:librechat:modelspec, librechat:librechat:agents, librechat:librechat:ocr ### MOD-02: Model freshness Judgment rule: Measure the days from a provider's general release to availability in the product for the last three major releases from OpenAI, Anthropic and Google (and Mistral where offered). Use dated changelogs or observation. Record deprecation notice for retired models. #### Dust Verification status: Agent cross-checked (agent-cross-checked) Support: partial The public changelog and exact agent model selection are documented; release-lag and removal notice remain unknown. Availability: generally-available Delivery: native Scope: surfaces=Dust web workspace, Dust EU web workspace; version=Public source pass 2026-10-21; no logged-in deployment observed; plan=Business credit-priced plan; Enterprise or legacy differences are named per answer; add_ons=none stated; deployment_mode=multi-tenant-saas; region=US/global or EU workspace as named; EU-hosted-model restriction must be enabled separately; os=not specified. Limitations: - Documented findings only; unestablished atomic questions remain explicit gaps. - release_lag_days — What was the lag for each of the last three major releases, per provider? [unknown]: What was the lag for each of the last three major releases, per provider? The dated changelog and model-selection docs were checked; last-three-release/provider lag measurements and a fixed advance deprecation-notice period were not established. - changelog — Is there a dated public model changelog? [supported]: The public changelog contains dated model releases including Gemini 4 Argon on October 14, 2026. Sources: dust:dust:changelog. - deprecation_notice — How much notice is given before a model is removed? [unknown]: How much notice is given before a model is removed? The dated changelog and model-selection docs were checked; last-three-release/provider lag measurements and a fixed advance deprecation-notice period were not established. - version_pinning — Can an admin keep an older model version? [partial]: An agent can select a specific model and keeps it until changed, subject to continued provider/workspace availability. No indefinite old-version entitlement is established. Sources: dust:dust:models. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: dust:dust:changelog, dust:dust:models #### LibreChat Verification status: Agent cross-checked (agent-cross-checked) Support: partial Dated releases document model updates, but measured availability lags and retirement notice periods remain unknown. Availability: generally-available Delivery: native Scope: surfaces=Customer-operated web application and server; version=Public documentation retrieved 2026-10-21; inspected source v0.8.8 / e8f3be08623663d4ad7f7241e693c94469b63bb0; plan=MIT customer-operated LibreChat; optional separately deployed services identified per answer; add_ons=none stated; deployment_mode=on-premise; region=Customer-selected infrastructure; model, OCR, search, observability and support services separately scoped; os=not specified. Limitations: - release_lag_days: What was the lag for each of the last three major releases, per provider? No paired general-release/LibreChat-availability dates for the last three major releases per provider, or binding retirement-notice period, were established. - deprecation_notice: How much notice is given before a model is removed? No paired general-release/LibreChat-availability dates for the last three major releases per provider, or binding retirement-notice period, were established. - version_pinning: Admins can name explicit model IDs in endpoints/model specs; continuing remote availability remains under provider control. - release_lag_days — What was the lag for each of the last three major releases, per provider? [unknown]: release_lag_days: What was the lag for each of the last three major releases, per provider? No paired general-release/LibreChat-availability dates for the last three major releases per provider, or binding retirement-notice period, were established. - changelog — Is there a dated public model changelog? [supported]: Public releases and the LibreChat changelog date model updates; v0.8.8 was published 2026-10-01. Sources: librechat:librechat:releases. - deprecation_notice — How much notice is given before a model is removed? [unknown]: deprecation_notice: How much notice is given before a model is removed? No paired general-release/LibreChat-availability dates for the last three major releases per provider, or binding retirement-notice period, were established. - version_pinning — Can an admin keep an older model version? [partial]: Admins can name explicit model IDs in endpoints/model specs; continuing remote availability remains under provider control. Sources: librechat:librechat:modelspec, librechat:librechat:custom. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: librechat:librechat:releases, librechat:librechat:modelspec, librechat:librechat:custom ### MOD-03: Bring-your-own keys and provider contracts Judgment rule: Record customer-held API keys, customer cloud accounts (Azure OpenAI, Bedrock, Vertex), and aggregator routes (OpenRouter, LiteLLM) as separate routes. Record what still works without a vendor-provided model, whether prices change with BYO, and documented restrictions. API-key support does not imply support for the customer's own Azure contract. #### Dust Verification status: No public information found (not-published) Support: not-published We found no public information on this. A missing finding is not evidence that the product lacks the capability. #### LibreChat Verification status: Agent cross-checked (agent-cross-checked) Support: partial Customer-held provider keys, customer Azure/AWS/Vertex configurations and custom endpoints are documented. Availability: generally-available Delivery: native Scope: surfaces=Customer-operated web application and server; version=Public documentation retrieved 2026-10-21; inspected source v0.8.8 / e8f3be08623663d4ad7f7241e693c94469b63bb0; plan=MIT customer-operated LibreChat; optional separately deployed services identified per answer; add_ons=none stated; deployment_mode=on-premise; region=Customer-selected infrastructure; model, OCR, search, observability and support services separately scoped; os=not specified. Limitations: - byo_restrictions: File modalities, tools and other capabilities depend on provider/model support and optional services. A universal BYO feature-parity guarantee was not established. - own_api_keys — Can the customer use its own provider API keys? [supported]: Provider keys can be operator-supplied or user_provided, with user key controls in Settings. Sources: librechat:librechat:firstchat, librechat:librechat:custom. - own_cloud_contract — Can it use its own Azure OpenAI, Bedrock or Vertex deployment? [supported]: Customer Azure OpenAI deployments, AWS Bedrock region/credentials and Anthropic Vertex project/region routes are documented separately. Sources: librechat:librechat:azure, librechat:librechat:bedrock, librechat:librechat:vertex. - custom_endpoint — Can an admin add any OpenAI-compatible endpoint? [supported]: Operators can configure a custom compatible API URL and model list. Sources: librechat:librechat:custom. - byo_pricing — Does BYO change the licence price? [supported]: MIT repository use has no per-seat licence fee; customers separately procure provider service usage. Sources: librechat:librechat:license, librechat:librechat:firstchat. - byo_restrictions — Which features stop working with BYO? [partial]: File modalities, tools and other capabilities depend on provider/model support and optional services. A universal BYO feature-parity guarantee was not established. Sources: librechat:librechat:ocr, librechat:librechat:code. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: librechat:librechat:firstchat, librechat:librechat:custom, librechat:librechat:azure, librechat:librechat:bedrock, librechat:librechat:vertex, librechat:librechat:license, librechat:librechat:ocr, librechat:librechat:code ### MOD-04: Local and self-hosted models Judgment rule: Record support for models running in the customer's environment (vLLM, Ollama, TGI or similar) and which features work with them: chat, RAG, tools, embeddings. Distinguish a documented, supported route from a generic endpoint that happens to work. #### Dust Verification status: No public information found (not-published) Support: not-published We found no public information on this. A missing finding is not evidence that the product lacks the capability. #### LibreChat Verification status: Agent cross-checked (agent-cross-checked) Support: partial Local compatible inference and Ollama embeddings are documented; optional services must also be local for a fully private stack. Availability: generally-available Delivery: native Scope: surfaces=Customer-operated web application and server; version=Public documentation retrieved 2026-10-21; inspected source v0.8.8 / e8f3be08623663d4ad7f7241e693c94469b63bb0; plan=MIT customer-operated LibreChat; optional separately deployed services identified per answer; add_ons=none stated; deployment_mode=on-premise; region=Customer-selected infrastructure; model, OCR, search, observability and support services separately scoped; os=not specified. Limitations: - feature_parity: Chat and retrieval can use local endpoints. Tool calling and vision depend on the model; external OCR/search/code services are separate. No complete offline fixture was run. - vendor_hosted_open_models: Does the vendor host open-weight models in EU/CH? A vendor-hosted EU/Swiss open-weight model service was not established for the MIT self-hosted project. - self_hosted_chat — Can chat use a customer-hosted model? [supported]: Custom endpoints can point at customer-hosted compatible model servers, including Ollama-compatible setups. Sources: librechat:librechat:custom. - self_hosted_embeddings — Can retrieval use customer-hosted embeddings? [supported]: The RAG configuration documents Ollama embeddings and local Hugging Face through the appropriate full image. Sources: librechat:librechat:ragcfg. - feature_parity — Which features fail with self-hosted models? [partial]: Chat and retrieval can use local endpoints. Tool calling and vision depend on the model; external OCR/search/code services are separate. No complete offline fixture was run. Sources: librechat:librechat:ragcfg, librechat:librechat:ocr, librechat:librechat:code. - vendor_hosted_open_models — Does the vendor host open-weight models in EU/CH? [unknown]: vendor_hosted_open_models: Does the vendor host open-weight models in EU/CH? A vendor-hosted EU/Swiss open-weight model service was not established for the MIT self-hosted project. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: librechat:librechat:custom, librechat:librechat:ragcfg, librechat:librechat:ocr, librechat:librechat:code ### MOD-05: Model control and usage limits Judgment rule: Record admin allow-lists per group, default models, restrictions on which models may see which data, and per-user rate or fair-use limits. Test one restriction with two users in different groups. Record limits as enforced stops, warnings or undocumented fair use. #### Dust Verification status: Agent cross-checked (agent-cross-checked) Support: partial Admins cap models by workspace, group and member, with explicit published-agent exceptions and credit limits. Availability: generally-available Delivery: native Scope: surfaces=Dust web workspace, Dust EU web workspace; version=Public source pass 2026-10-21; no logged-in deployment observed; plan=Business credit-priced plan; Enterprise or legacy differences are named per answer; add_ons=none stated; deployment_mode=multi-tenant-saas; region=US/global or EU workspace as named; EU-hosted-model restriction must be enabled separately; os=not specified. Limitations: - No two-user enforcement test. Published agents can run above member tiers if the separate admin setting permits it. - model_allowlist_by_group — Can admins limit models per group? [supported]: Model access ceilings apply to the workspace, groups or individuals and cover reasoning efforts as well as auto selections. Sources: dust:dust:tiers. - data_bound_models — Can admins restrict which models receive connector or knowledge content? [partial]: EU-only mode restricts all active agents and pickers to EU-hosted models. Spaces bound knowledge and agent access; arbitrary per-data-class model rules were not established. Sources: dust:dust:eu-models, dust:dust:spaces. - default_model — Can admins set the default model? [supported]: Builders configure an agent default, which may be exact or Basic/Standard/Premium. Composer changes do not edit it for others. Sources: dust:dust:models. - user_limits — What per-user message or token limits apply, and are they published? [supported]: Free seats have 500 lifetime credits, Pro 8,000/month and Max 40,000/month, plus applicable shared-pool rules and member spend limits. Sources: dust:dust:credits, dust:dust:credit-management. - limit_behaviour — What happens when a limit is reached? [supported]: Members block when applicable sources/caps are exhausted. A message begun under a cap can finish above it; new messages then block. A separate root-task checkpoint can pause at 600 AWU. Sources: dust:dust:credit-management, dust:dust:credits. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: dust:dust:tiers, dust:dust:eu-models, dust:dust:spaces, dust:dust:models, dust:dust:credits, dust:dust:credit-management #### LibreChat Verification status: Agent cross-checked (agent-cross-checked) Support: partial Model specs and scoped configuration can limit choices; optional user credit reservations provide a blocking mechanism. Availability: generally-available Delivery: native Scope: surfaces=Customer-operated web application and server; version=Public documentation retrieved 2026-10-21; inspected source v0.8.8 / e8f3be08623663d4ad7f7241e693c94469b63bb0; plan=MIT customer-operated LibreChat; optional separately deployed services identified per answer; add_ons=none stated; deployment_mode=on-premise; region=Customer-selected infrastructure; model, OCR, search, observability and support services separately scoped; os=not specified. Limitations: - model_allowlist_by_group: Enforced model specs and role/group configuration overrides can restrict available endpoints/models. The separate management UI is preview; a two-group fixture was not run. - data_bound_models: Can admins restrict which models receive connector or knowledge content? A hard binding between each enterprise content source and a permitted model provider was not established in the Agent and access-control guides. - user_limits: Balance tracking, initial credits/refills and message moderation limits are operator-configurable, rather than a universal vendor allowance. - model_allowlist_by_group — Can admins limit models per group? [partial]: Enforced model specs and role/group configuration overrides can restrict available endpoints/models. The separate management UI is preview; a two-group fixture was not run. Sources: librechat:librechat:modelspec, librechat:librechat:admin, librechat:librechat:access. - data_bound_models — Can admins restrict which models receive connector or knowledge content? [unknown]: data_bound_models: Can admins restrict which models receive connector or knowledge content? A hard binding between each enterprise content source and a permitted model provider was not established in the Agent and access-control guides. - default_model — Can admins set the default model? [supported]: Model-spec presets can define default endpoint/model configurations. Sources: librechat:librechat:modelspec. - user_limits — What per-user message or token limits apply, and are they published? [partial]: Balance tracking, initial credits/refills and message moderation limits are operator-configurable, rather than a universal vendor allowance. Sources: librechat:librechat:balance, librechat:librechat:mod. - limit_behaviour — What happens when a limit is reached? [supported]: Pinned checkBalance reserves credits and throws a TOKEN_BALANCE error when insufficient. This establishes a blocking code path, not an observed end-to-end quota test. Sources: librechat:librechat:balance-pin. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: librechat:librechat:modelspec, librechat:librechat:admin, librechat:librechat:access, librechat:librechat:balance, librechat:librechat:mod, librechat:librechat:balance-pin ## Company knowledge ### KNOW-01: Files in chat Judgment rule: Record supported formats, size and count limits, scanned-PDF OCR, tables and spreadsheets, images inside documents, and whether a file is read in full or retrieved in chunks. Upload the shared fixture set (long German PDF, scanned contract, Excel with formulas, PowerPoint) and record failures. #### Dust Verification status: Agent cross-checked (agent-cross-checked) Support: partial Conversation files and code/OCR handling are documented; conflicting effective upload caps are explicitly unknown. Availability: generally-available Delivery: native Scope: surfaces=Dust web workspace, Dust EU web workspace; version=Public source pass 2026-10-21; no logged-in deployment observed; plan=Business credit-priced plan; Enterprise or legacy differences are named per answer; add_ons=none stated; deployment_mode=multi-tenant-saas; region=US/global or EU workspace as named; EU-hosted-model restriction must be enabled separately; os=not specified. Limitations: - No common PDF/scan/Excel/PowerPoint fixture run. - Do not use either disputed cap set as the effective Business upload limit. - formats — Which formats are accepted? [supported]: Conversation uploads accept documents/PDF, text/code, Excel/CSV/TSV, images and supported audio. Direct video uploads are explicitly excluded by the limits FAQ. Sources: dust:dust:files, dust:dust:file-limits. - limits — What size and count limits apply per file and chat? [unknown]: Effective image/audio/tabular maximums and file counts are unknown: Conversation Files lists5/25/50MB while File size limits lists20/100/350MB. Both mention50MB general data files, but neither resolves the actual current per-plan upload route. No logged-in upload test was performed. - ocr — Are scanned PDFs read? [partial]: Computer documents OCR for English, French and mixed scans with page-level caveats. Drive connector itself ignores image-only files; upload to a conversation is the documented route. Sources: dust:dust:computer, dust:dust:drive. - spreadsheets — Are Excel tables read with structure intact? [partial]: Computer can read, calculate and edit uploaded Excel with code. Formula/style preservation must be checked; no fixture fidelity result is available. Sources: dust:dust:computer. - full_vs_chunked — Is the whole file used or only retrieved parts, and does the user see which? [partial]: Conversation uploads are not indexed in workspace sources. Computer can inspect full files, while knowledge search retrieves relevant context; complete chunk visibility was not verified. Sources: dust:dust:files, dust:dust:computer, dust:dust:search. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: dust:dust:files, dust:dust:file-limits, dust:dust:computer, dust:dust:drive, dust:dust:search #### LibreChat Verification status: Agent cross-checked (agent-cross-checked) Support: partial Files can be indexed, injected as extracted text or sent directly to compatible model providers. Availability: generally-available Delivery: native Scope: surfaces=Customer-operated web application and server; version=New spreadsheet finding at development source a378657583567310c841d652b0f6b00c83e2bd6b; earlier accepted facts retain their cited release pins; plan=MIT customer-operated LibreChat; optional separately deployed services identified per answer; add_ons=none stated; deployment_mode=on-premise; region=Customer-selected infrastructure; model, OCR, search, observability and support services separately scoped; os=not specified. Limitations: - File counts, sizes and provider acceptance depend on deployment settings. - Spreadsheet parsing yields named sheets and CSV values; layout/formula fidelity and the shared fixture remain untested. - formats — Which formats are accepted? [supported]: Documents include PDF/text/Office formats through configured extraction; provider-native acceptance varies, including Bedrock Office formats and Google audio/video. Sources: librechat:librechat:ocr, librechat:librechat:filecfg. - limits — What size and count limits apply per file and chat? [partial]: File counts, sizes and MIME patterns are operator/endpoint settings; provider limits may be stricter. The shared fixture was not uploaded. Sources: librechat:librechat:filecfg, librechat:librechat:ocr. - ocr — Are scanned PDFs read? [supported]: Optional Mistral OCR, Azure/Vertex Mistral OCR and provider-native file processing are documented. Sources: librechat:librechat:ocr. - spreadsheets — Are Excel tables read with structure intact? [partial]: The inspected built-in XLSX/XLS/ODS parser emits each sheet name and CSV values. The current guide lists a 15 MB built-in parser limit. Workbook formatting, formula recalculation and shared-fixture accuracy were not established. Sources: librechat:librechat:r4-spreadsheet-parser, librechat:librechat:r4-text-current. - full_vs_chunked — Is the whole file used or only retrieved parts, and does the user see which? [supported]: RAG retrieves indexed chunks; Upload as Text and Agent File Context inject extracted full text, subject to context limits. Sources: librechat:librechat:rag, librechat:librechat:text, librechat:librechat:ocr. Checked 2026-10-26; agent cross-checked 2026-10-26; check again by 2026-12-25. Sources: librechat:librechat:ocr, librechat:librechat:filecfg, librechat:librechat:rag, librechat:librechat:text, librechat:librechat:r4-spreadsheet-parser, librechat:librechat:r4-text-current ### KNOW-02: Knowledge bases and shared collections Judgment rule: Record curated document collections that persist across chats, who can create and share them, size limits, and how they attach to assistants. Distinguish personal, group and organisation-wide scope. #### Dust Verification status: Agent cross-checked (agent-cross-checked) Support: partial Spaces persist shared knowledge and bind it to agent tools. Availability: generally-available Delivery: native Scope: surfaces=Dust web workspace, Dust EU web workspace; version=Public source pass 2026-10-21; no logged-in deployment observed; plan=Business credit-priced plan; Enterprise or legacy differences are named per answer; add_ons=none stated; deployment_mode=multi-tenant-saas; region=US/global or EU workspace as named; EU-hosted-model restriction must be enabled separately; os=not specified. Limitations: - Documented findings only; unestablished atomic questions remain explicit gaps. - persistent_collections — Can users keep document collections across chats? [supported]: Open and restricted Spaces retain connected sources, folders and other data across conversations. Sources: dust:dust:spaces. - sharing_scope — Can collections be shared with users, groups or the whole organisation? [supported]: Open Spaces are workspace-wide; restricted Spaces have explicit members or provisioned groups. Sources: dust:dust:spaces. - size_limits — What are the collection limits? [partial]: Business supports five Spaces and Enterprise 100. Complete per-Space storage/document quotas were not established. Sources: dust:dust:spaces. - assistant_binding — Can collections be bound to an assistant? [supported]: An agent tool can use selected Space data; Space membership governs agent visibility and use. Sources: dust:dust:spaces. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: dust:dust:spaces #### LibreChat Verification status: Agent cross-checked (agent-cross-checked) Support: partial Agents and personal Projects keep files across chats; shared Agent access is distinct from personal Project ownership. Availability: generally-available Delivery: native Scope: surfaces=Customer-operated web application and server; version=Public documentation retrieved 2026-10-21; inspected source v0.8.8 / e8f3be08623663d4ad7f7241e693c94469b63bb0; plan=MIT customer-operated LibreChat; optional separately deployed services identified per answer; add_ons=none stated; deployment_mode=on-premise; region=Customer-selected infrastructure; model, OCR, search, observability and support services separately scoped; os=not specified. Limitations: - sharing_scope: Agents can share files/knowledge with permitted users/groups/roles; Projects themselves are personal and cannot be shared or transferred. - size_limits: What are the collection limits? No universal collection storage/size allowance was established; limits are deployment and file-configuration dependent. - persistent_collections — Can users keep document collections across chats? [supported]: Agent File Search/File Context and Project reference files persist for later chats. Sources: librechat:librechat:agents, librechat:librechat:projects. - sharing_scope — Can collections be shared with users, groups or the whole organisation? [partial]: Agents can share files/knowledge with permitted users/groups/roles; Projects themselves are personal and cannot be shared or transferred. Sources: librechat:librechat:access, librechat:librechat:agents, librechat:librechat:projects. - size_limits — What are the collection limits? [unknown]: size_limits: What are the collection limits? No universal collection storage/size allowance was established; limits are deployment and file-configuration dependent. - assistant_binding — Can collections be bound to an assistant? [supported]: Agent File Search and File Context bind retrieved or extracted knowledge to an Agent. Sources: librechat:librechat:agents, librechat:librechat:ocr. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: librechat:librechat:agents, librechat:librechat:projects, librechat:librechat:access, librechat:librechat:ocr ### KNOW-03: Enterprise connectors Judgment rule: List each connector separately with plan, availability and setup owner: SharePoint, OneDrive, Outlook, Teams, Google Drive, Gmail, Confluence, Jira, Notion, Box, Salesforce, ServiceNow, file shares. Record whether the vendor or a third party (or the customer via MCP) provides it, and whether it needs admin consent. A connector listed on a roadmap or partner page is `announced`. An on-prem connector or gateway proves access to local data, not a self-hosted platform (DATA-02). #### Dust Verification status: Agent cross-checked (agent-cross-checked) Support: partial Managed content connectors coexist with MCP tools and beta imports; Microsoft content indexing is account-scoped. Availability: generally-available Delivery: official-integration Scope: surfaces=Dust web workspace, Dust EU web workspace; version=Public source pass 2026-10-21; no logged-in deployment observed; plan=Business: up to3 connections and5 remote MCP servers; Enterprise: unlimited connection/MCP packaging; beta imports identified separately; add_ons=none stated; deployment_mode=multi-tenant-saas; region=US/global or EU workspace as named; EU-hosted-model restriction must be enabled separately; os=not specified. Limitations: - Documented findings only; unestablished atomic questions remain explicit gaps. - Managed sources, metadata/table-query connections, MCP tools and beta imports are separate routes; no all-systems managed-index promise. - microsoft_365 — Which M365 sources (SharePoint, OneDrive, Outlook, Teams) are connected? [partial]: The Microsoft connection syncs selected SharePoint drives/documents within the connecting account permissions. Teams-associated SharePoint content is described; this does not establish Outlook mail or all Teams messages as indexed sources. Sources: dust:dust:microsoft. - google_workspace — Which Google sources are connected? [partial]: Drive documents, Sheets, Slides and text PDFs can sync. Gmail is separately available as an MCP tool in the credits/tool catalog, not established as the same managed content index. Sources: dust:dust:drive, dust:dust:credits. - atlassian — Confluence and Jira? [partial]: Confluence indexes selected global spaces, excluding restricted pages. Jira has a beta custom import and MCP tool route; do not count the beta import as a general managed connector. Sources: dust:dust:confluence, dust:dust:jira, dust:dust:credits. - other_systems — Which other business systems? [partial]: Documented routes include managed Slack/Notion/GitHub and Salesforce/Zendesk/Snowflake sources/tools. BigQuery specifically uses a dedicated-service-account table-query connection: Dust keeps metadata, not a synchronized table-content index; queries stay within the chosen BigQuery dataset location and Space admins select tables. General Business/Enterprise connection limits apply; a BigQuery-specific commercial add-on is not separately specified. Sources: dust:dust:sources, dust:dust:credits, dust:dust:r2-bigquery, dust:dust:pricing. - on_prem_sources — Can on-prem file shares or databases be connected? [partial]: Admins can connect a customer-hosted MCP server at a reachable secured URL. This is tool access, not evidence of a native SMB/file-share connector or on-premises Dust deployment. Sources: dust:dust:tools. - connector_provider — Who builds and maintains each connector? [supported]: Dust manages built-in connections; external MCP servers can be vendor/community/customer hosted and authenticated separately. Sources: dust:dust:permissions, dust:dust:tools. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: dust:dust:microsoft, dust:dust:drive, dust:dust:credits, dust:dust:confluence, dust:dust:jira, dust:dust:sources, dust:dust:r2-bigquery, dust:dust:pricing, dust:dust:tools, dust:dust:permissions #### LibreChat Verification status: Agent cross-checked (agent-cross-checked) Support: partial SharePoint and OneDrive have a documented delegated file picker. Other enterprise systems require separately verified tools or connectors. Availability: generally-available Delivery: native Scope: surfaces=Customer-operated web application and server; version=Public documentation retrieved 2026-10-21; inspected source v0.8.8 / e8f3be08623663d4ad7f7241e693c94469b63bb0; plan=MIT customer-operated LibreChat; optional separately deployed services identified per answer; add_ons=none stated; deployment_mode=on-premise; region=Customer-selected infrastructure; model, OCR, search, observability and support services separately scoped; os=not specified. Limitations: - microsoft_365: SharePoint Online and OneDrive for Business can be browsed and attached through Entra token reuse/OBO, delegated AllSites.Read and Files.Read.All, and admin consent. Outlook/Teams source access was not established. - google_workspace: Which Google sources are connected? Google Drive/Gmail, Confluence/Jira and other named business-system connectors were not verified from first-party maintained connector documentation; generic MCP compatibility is not a connector catalog. - atlassian: Confluence and Jira? Google Drive/Gmail, Confluence/Jira and other named business-system connectors were not verified from first-party maintained connector documentation; generic MCP compatibility is not a connector catalog. - other_systems: Which other business systems? Google Drive/Gmail, Confluence/Jira and other named business-system connectors were not verified from first-party maintained connector documentation; generic MCP compatibility is not a connector catalog. - on_prem_sources: Custom local MCP stdio servers can reach systems available to the deployment, but a maintained file-share/database connector and its permissions were not verified. - microsoft_365 — Which M365 sources (SharePoint, OneDrive, Outlook, Teams) are connected? [partial]: SharePoint Online and OneDrive for Business can be browsed and attached through Entra token reuse/OBO, delegated AllSites.Read and Files.Read.All, and admin consent. Outlook/Teams source access was not established. Sources: librechat:librechat:sharepoint. - google_workspace — Which Google sources are connected? [unknown]: google_workspace: Which Google sources are connected? Google Drive/Gmail, Confluence/Jira and other named business-system connectors were not verified from first-party maintained connector documentation; generic MCP compatibility is not a connector catalog. - atlassian — Confluence and Jira? [unknown]: atlassian: Confluence and Jira? Google Drive/Gmail, Confluence/Jira and other named business-system connectors were not verified from first-party maintained connector documentation; generic MCP compatibility is not a connector catalog. - other_systems — Which other business systems? [unknown]: other_systems: Which other business systems? Google Drive/Gmail, Confluence/Jira and other named business-system connectors were not verified from first-party maintained connector documentation; generic MCP compatibility is not a connector catalog. - on_prem_sources — Can on-prem file shares or databases be connected? [partial]: Custom local MCP stdio servers can reach systems available to the deployment, but a maintained file-share/database connector and its permissions were not verified. Sources: librechat:librechat:mcpcfg. - connector_provider — Who builds and maintains each connector? [supported]: LibreChat documents the Microsoft picker. Other MCP tool integrations are separately configured and maintained by their respective server authors/operators. Sources: librechat:librechat:sharepoint, librechat:librechat:mcp. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: librechat:librechat:sharepoint, librechat:librechat:mcpcfg, librechat:librechat:mcp ### KNOW-04: Permission-aware retrieval Judgment rule: Using two test users, verify that a restricted document in the fixture tenant is never cited or summarised for the user without access, including after a permission change. Record whether permissions are enforced at query time by the source system, mirrored into an index, or not enforced. Record the propagation delay of a revocation. #### Dust Verification status: Agent cross-checked (agent-cross-checked) Support: partial Knowledge access is documented through connection-account scope and Dust Spaces; per-user source-ACL mirroring is unverified. Availability: generally-available Delivery: native Scope: surfaces=Dust web workspace, Dust EU web workspace; version=Public source pass 2026-10-21; no logged-in deployment observed; plan=Business credit-priced plan; Enterprise or legacy differences are named per answer; add_ons=none stated; deployment_mode=multi-tenant-saas; region=US/global or EU workspace as named; EU-hosted-model restriction must be enabled separately; os=not specified. Limitations: - Connector access through one account is not counted as demonstrated per-user source permission preservation. - source_permissions_respected — Are source permissions enforced per user? [partial]: Drive and Microsoft content available to Dust follows the connecting account and admin-selected data. Users then access agent knowledge through Space membership. Individual source ACL inheritance for each querying user was not established. Sources: dust:dust:drive, dust:dust:microsoft, dust:dust:permissions. - enforcement_point — Does the source system, a mirrored index or nothing enforce them? [supported]: Managed content is synchronized under the connection identity and gated through open/restricted Spaces; personal tool credentials are a separate per-user route. Sources: dust:dust:permissions, dust:dust:credentials. - revocation_delay — How quickly does a removed permission take effect? [unknown]: How quickly does a removed permission take effect? No two-user restricted-document or revocation fixture was run. Checked connector/Space docs did not establish per-query source ACL mirrors, a source-permission revocation SLA, or Microsoft Purview-label enforcement. - admin_scope_controls — Can admins exclude sites, folders or labels? [supported]: Admins choose connector folders/documents/sites and control restricted Space membership. Sources: dust:dust:spaces, dust:dust:microsoft. - sensitivity_labels — Are Microsoft Purview or similar labels respected? [unknown]: Are Microsoft Purview or similar labels respected? No two-user restricted-document or revocation fixture was run. Checked connector/Space docs did not establish per-query source ACL mirrors, a source-permission revocation SLA, or Microsoft Purview-label enforcement. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: dust:dust:drive, dust:dust:microsoft, dust:dust:permissions, dust:dust:credentials, dust:dust:spaces #### LibreChat Verification status: Agent cross-checked (agent-cross-checked) Support: partial Product resource ACLs and delegated Microsoft file selection are documented. Continuing source permissions for copied content remain untested. Availability: generally-available Delivery: native Scope: surfaces=Customer-operated web application and server; version=Public documentation retrieved 2026-10-21; inspected source v0.8.8 / e8f3be08623663d4ad7f7241e693c94469b63bb0; plan=MIT customer-operated LibreChat; optional separately deployed services identified per answer; add_ons=none stated; deployment_mode=on-premise; region=Customer-selected infrastructure; model, OCR, search, observability and support services separately scoped; os=not specified. Limitations: - source_permissions_respected: Are source permissions enforced per user? No restricted-document two-user fixture or grant/revocation test was run. Source permission enforcement, revocation delay and Purview label handling remain unverified. - enforcement_point: LibreChat ACLs control Agent/file resources; Microsoft delegated OBO limits source downloads as the signed-in user. No query-time source ACL check for previously copied chunks was established. - revocation_delay: How quickly does a removed permission take effect? No restricted-document two-user fixture or grant/revocation test was run. Source permission enforcement, revocation delay and Purview label handling remain unverified. - admin_scope_controls: Operators can scope Agent/resource grants and its selected files. Site/folder/label exclusion across sources was not established. - sensitivity_labels: Are Microsoft Purview or similar labels respected? No restricted-document two-user fixture or grant/revocation test was run. Source permission enforcement, revocation delay and Purview label handling remain unverified. - source_permissions_respected — Are source permissions enforced per user? [unknown]: source_permissions_respected: Are source permissions enforced per user? No restricted-document two-user fixture or grant/revocation test was run. Source permission enforcement, revocation delay and Purview label handling remain unverified. - enforcement_point — Does the source system, a mirrored index or nothing enforce them? [partial]: LibreChat ACLs control Agent/file resources; Microsoft delegated OBO limits source downloads as the signed-in user. No query-time source ACL check for previously copied chunks was established. Sources: librechat:librechat:access, librechat:librechat:sharepoint. - revocation_delay — How quickly does a removed permission take effect? [unknown]: revocation_delay: How quickly does a removed permission take effect? No restricted-document two-user fixture or grant/revocation test was run. Source permission enforcement, revocation delay and Purview label handling remain unverified. - admin_scope_controls — Can admins exclude sites, folders or labels? [partial]: Operators can scope Agent/resource grants and its selected files. Site/folder/label exclusion across sources was not established. Sources: librechat:librechat:access, librechat:librechat:agents. - sensitivity_labels — Are Microsoft Purview or similar labels respected? [unknown]: sensitivity_labels: Are Microsoft Purview or similar labels respected? No restricted-document two-user fixture or grant/revocation test was run. Source permission enforcement, revocation delay and Purview label handling remain unverified. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: librechat:librechat:access, librechat:librechat:sharepoint, librechat:librechat:agents ### KNOW-05: Freshness and sync model Judgment rule: Record live (federated) search versus a copied index, sync interval, deletion propagation, and where the index is stored. Edit and delete one fixture document and record when answers change. A copied index adds storage location to DATA-01. #### Dust Verification status: Agent cross-checked (agent-cross-checked) Support: partial Managed connectors copy and synchronize content, with source-specific refresh and deletion timing. Availability: generally-available Delivery: native Scope: surfaces=Dust web workspace, Dust EU web workspace; version=Public source pass 2026-10-21; no logged-in deployment observed; plan=Business credit-priced plan; Enterprise or legacy differences are named per answer; add_ons=none stated; deployment_mode=multi-tenant-saas; region=US/global or EU workspace as named; EU-hosted-model restriction must be enabled separately; os=not specified. Limitations: - Documented findings only; unestablished atomic questions remain explicit gaps. - live_or_indexed — Is content searched live or copied into an index? [supported]: Connections synchronize content into Dust; conversation uploads remain scoped to the chat. MCP tools can access live services separately. Sources: dust:dust:permissions, dust:dust:files, dust:dust:tools. - sync_interval — How often does an index refresh? [supported]: Microsoft describes five-minute sync cycles; Drive says a few minutes; Confluence describes event updates plus an approximate hourly check. Sources: dust:dust:microsoft, dust:dust:drive, dust:dust:confluence. - deletion_propagation — How quickly does a deleted document disappear? [partial]: Microsoft changes/deletions appear on a subsequent sync and large batches can take longer. Drive deletions occur during its sync. No measured end-to-end bound was tested. Sources: dust:dust:microsoft, dust:dust:drive. - index_location — Where is indexed content stored? [partial]: Dust markets EU or US hosting; exact per-index country, cloud provider and backup location were not established. Sources: dust:dust:security. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: dust:dust:permissions, dust:dust:files, dust:dust:tools, dust:dust:microsoft, dust:dust:drive, dust:dust:confluence, dust:dust:security #### LibreChat Verification status: Agent cross-checked (agent-cross-checked) Support: partial The RAG API copies uploaded content into PGVector; imported files do not by themselves establish live enterprise synchronization. Availability: generally-available Delivery: native Scope: surfaces=Customer-operated web application and server; version=Public documentation retrieved 2026-10-21; inspected source v0.8.8 / e8f3be08623663d4ad7f7241e693c94469b63bb0; plan=MIT customer-operated LibreChat; optional separately deployed services identified per answer; add_ons=none stated; deployment_mode=on-premise; region=Customer-selected infrastructure; model, OCR, search, observability and support services separately scoped; os=not specified. Limitations: - sync_interval: How often does an index refresh? No periodic enterprise-source sync contract, source edit/delete propagation bound or timed fixture was established from the RAG and Microsoft import guides. - deletion_propagation: How quickly does a deleted document disappear? No periodic enterprise-source sync contract, source edit/delete propagation bound or timed fixture was established from the RAG and Microsoft import guides. - live_or_indexed — Is content searched live or copied into an index? [supported]: The RAG API indexes uploaded documents into PostgreSQL/PGVector; file-context mode stores extracted text in Agent instructions. Sources: librechat:librechat:rag, librechat:librechat:ocr. - sync_interval — How often does an index refresh? [unknown]: sync_interval: How often does an index refresh? No periodic enterprise-source sync contract, source edit/delete propagation bound or timed fixture was established from the RAG and Microsoft import guides. - deletion_propagation — How quickly does a deleted document disappear? [unknown]: deletion_propagation: How quickly does a deleted document disappear? No periodic enterprise-source sync contract, source edit/delete propagation bound or timed fixture was established from the RAG and Microsoft import guides. - index_location — Where is indexed content stored? [supported]: The operator chooses the RAG database and its location, independently of the inference provider. Sources: librechat:librechat:rag, librechat:librechat:compose-pin. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: librechat:librechat:rag, librechat:librechat:ocr, librechat:librechat:compose-pin ### KNOW-06: Grounding and retrieval quality Judgment rule: Run the shared question set over the fixture corpus: factual lookups, cross-document questions, German document with an English question, a question with no answer in the corpus. Record citation presence, link accuracy, correct answers and refusals to invent. Disclose model, settings, attempt count and failures. Single runs support a case study, not a ranking. #### Dust Verification status: Agent cross-checked (agent-cross-checked) Support: partial Knowledge and research tools cite sources; observed retrieval quality remains unknown. Availability: generally-available Delivery: native Scope: surfaces=Dust web workspace, Dust EU web workspace; version=Public source pass 2026-10-21; no logged-in deployment observed; plan=Business credit-priced plan; Enterprise or legacy differences are named per answer; add_ons=none stated; deployment_mode=multi-tenant-saas; region=US/global or EU workspace as named; EU-hosted-model restriction must be enabled separately; os=not specified. Limitations: - Documented findings only; unestablished atomic questions remain explicit gaps. - citations — Are answers linked to specific sources? [supported]: Deep-dive documents citations to documents, web pages and data queries. Sources: dust:dust:deep. - citation_accuracy — Do the cited passages support the answer? [unknown]: Do the cited passages support the answer? No multilingual shared corpus or question set was run. Citation accuracy, cross-language retrieval success, no-answer behavior and correct-answer counts remain unmeasured; model multilingual ability is not a retrieval-quality test. - cross_language — Does a question in one language find documents in another? [unknown]: Does a question in one language find documents in another? No multilingual shared corpus or question set was run. Citation accuracy, cross-language retrieval success, no-answer behavior and correct-answer counts remain unmeasured; model multilingual ability is not a retrieval-quality test. - no_answer_behaviour — Does it say when the corpus has no answer? [unknown]: Does it say when the corpus has no answer? No multilingual shared corpus or question set was run. Citation accuracy, cross-language retrieval success, no-answer behavior and correct-answer counts remain unmeasured; model multilingual ability is not a retrieval-quality test. - observed_accuracy — How many of the question set were answered correctly? [unknown]: How many of the question set were answered correctly? No multilingual shared corpus or question set was run. Citation accuracy, cross-language retrieval success, no-answer behavior and correct-answer counts remain unmeasured; model multilingual ability is not a retrieval-quality test. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: dust:dust:deep #### LibreChat Verification status: Agent cross-checked (agent-cross-checked) Support: partial File-search citations are documented; retrieval accuracy remains unmeasured. Availability: generally-available Delivery: native Scope: surfaces=Customer-operated web application and server; version=Public documentation retrieved 2026-10-21; inspected source v0.8.8 / e8f3be08623663d4ad7f7241e693c94469b63bb0; plan=MIT customer-operated LibreChat; optional separately deployed services identified per answer; add_ons=none stated; deployment_mode=on-premise; region=Customer-selected infrastructure; model, OCR, search, observability and support services separately scoped; os=not specified. Limitations: - citation_accuracy: Do the cited passages support the answer? The shared multilingual factual/cross-document/no-answer question set was not run. Citation correctness, cross-language success, no-answer reliability and observed accuracy remain unknown. - cross_language: Does a question in one language find documents in another? The shared multilingual factual/cross-document/no-answer question set was not run. Citation correctness, cross-language success, no-answer reliability and observed accuracy remain unknown. - no_answer_behaviour: Does it say when the corpus has no answer? The shared multilingual factual/cross-document/no-answer question set was not run. Citation correctness, cross-language success, no-answer reliability and observed accuracy remain unknown. - observed_accuracy: How many of the question set were answered correctly? The shared multilingual factual/cross-document/no-answer question set was not run. Citation correctness, cross-language success, no-answer reliability and observed accuracy remain unknown. - citations — Are answers linked to specific sources? [supported]: RAG/File Search provides retrieved source references and citation controls. Sources: librechat:librechat:rag, librechat:librechat:access. - citation_accuracy — Do the cited passages support the answer? [unknown]: citation_accuracy: Do the cited passages support the answer? The shared multilingual factual/cross-document/no-answer question set was not run. Citation correctness, cross-language success, no-answer reliability and observed accuracy remain unknown. - cross_language — Does a question in one language find documents in another? [unknown]: cross_language: Does a question in one language find documents in another? The shared multilingual factual/cross-document/no-answer question set was not run. Citation correctness, cross-language success, no-answer reliability and observed accuracy remain unknown. - no_answer_behaviour — Does it say when the corpus has no answer? [unknown]: no_answer_behaviour: Does it say when the corpus has no answer? The shared multilingual factual/cross-document/no-answer question set was not run. Citation correctness, cross-language success, no-answer reliability and observed accuracy remain unknown. - observed_accuracy — How many of the question set were answered correctly? [unknown]: observed_accuracy: How many of the question set were answered correctly? The shared multilingual factual/cross-document/no-answer question set was not run. Citation correctness, cross-language success, no-answer reliability and observed accuracy remain unknown. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: librechat:librechat:rag, librechat:librechat:access ### KNOW-07: Memory and personal context Judgment rule: Record personal custom instructions, cross-chat memory, project or workspace context, and user controls to view, edit, delete or disable memory. Record whether admins can disable memory organisation-wide. #### Dust Verification status: Agent cross-checked (agent-cross-checked) Support: partial Agent-specific private memory persists across chats and can be viewed or deleted. Availability: generally-available Delivery: native Scope: surfaces=Dust web workspace, Dust EU web workspace; version=Public source pass 2026-10-21; no logged-in deployment observed; plan=Business credit-priced plan; Enterprise or legacy differences are named per answer; add_ons=none stated; deployment_mode=multi-tenant-saas; region=US/global or EU workspace as named; EU-hosted-model restriction must be enabled separately; os=not specified. Limitations: - Documented findings only; unestablished atomic questions remain explicit gaps. - custom_instructions — Can users set persistent instructions? [partial]: Agent instructions persist in the agent configuration and saved memory can contain user preferences; a separate personal global-instructions setting was not established. Sources: dust:dust:builder, dust:dust:memory. - cross_chat_memory — Does the product remember facts across chats? [supported]: The Memory tool retains user-specific information across conversations per agent. Sources: dust:dust:memory. - user_memory_controls — Can users view, edit and delete memories? [partial]: Users can view and delete agent memories, which persist until deletion. Direct editing or a global personal off switch was not established. Sources: dust:dust:memory. - admin_memory_controls — Can admins turn memory off? [unknown]: Can admins turn memory off? The Agent Memory and governance guides were checked; an admin-wide switch to disable all memory was not established. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: dust:dust:builder, dust:dust:memory #### LibreChat Verification status: Agent cross-checked (agent-cross-checked) Support: supported Opt-in user memory stores editable key/value context across chats; Projects and Agents provide additional persistent instructions. Availability: generally-available Delivery: native Scope: surfaces=Customer-operated web application and server; version=Public documentation retrieved 2026-10-21; inspected source v0.8.8 / e8f3be08623663d4ad7f7241e693c94469b63bb0; plan=MIT customer-operated LibreChat; optional separately deployed services identified per answer; add_ons=none stated; deployment_mode=on-premise; region=Customer-selected infrastructure; model, OCR, search, observability and support services separately scoped; os=not specified. - custom_instructions — Can users set persistent instructions? [supported]: Projects retain instructions across their chats; Agents retain system instructions. Sources: librechat:librechat:projects, librechat:librechat:agents. - cross_chat_memory — Does the product remember facts across chats? [supported]: User Memory injects configured key/value entries across chats and is not semantic search over all past conversations. Sources: librechat:librechat:memory. - user_memory_controls — Can users view, edit and delete memories? [supported]: Users add/edit/remove entries and can toggle memory when permitted. Sources: librechat:librechat:memory. - admin_memory_controls — Can admins turn memory off? [supported]: Memory is disabled until configured and its feature permissions can be restricted by role. Sources: librechat:librechat:memory, librechat:librechat:access. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: librechat:librechat:projects, librechat:librechat:agents, librechat:librechat:memory, librechat:librechat:access ## Assistants and agents ### AGT-01: Custom assistants Judgment rule: Record creation of assistants with instructions, model, knowledge and tools; who can create; sharing scope; versioning; and an internal catalogue or directory. Build one HR policy assistant on the fixture and share it with one group. Prompt templates and a prompt library are recorded here. #### Dust Verification status: Agent cross-checked (agent-cross-checked) Support: partial Custom agents have instructions, models, knowledge and tools with group-governed creation and sharing. Availability: generally-available Delivery: native Scope: surfaces=Dust web workspace, Dust EU web workspace; version=Public source pass 2026-10-21; no logged-in deployment observed; plan=Business credit-priced plan; Enterprise or legacy differences are named per answer; add_ons=none stated; deployment_mode=multi-tenant-saas; region=US/global or EU workspace as named; EU-hosted-model restriction must be enabled separately; os=not specified. Limitations: - Documented findings only; unestablished atomic questions remain explicit gaps. - create_assistants — Can users create assistants with instructions, knowledge and tools? [supported]: Agent Builder configures instructions, models, knowledge and tools. Sources: dust:dust:builder. - creator_permissions — Can admins limit who creates them? [supported]: Create/publish agent and skill permissions can be Everyone, Groups or Admin only. Sources: dust:dust:governance. - sharing_scope — Can assistants be shared with users, groups or everyone? [supported]: Agents and their knowledge access are controlled by sharing and Space memberships; group governance regulates publishing. Sources: dust:dust:builder, dust:dust:spaces, dust:dust:governance. - catalogue — Is there an internal assistant directory? [partial]: Published/shared agents and default agents are discoverable in the workspace; exact catalog configuration was not observed. Sources: dust:dust:builder, dust:dust:models. - prompt_library — Are shared prompt templates available? [partial]: Reusable agents and skills package instructions, knowledge and tools. A separate legacy prompt-library feature is not assumed. Sources: dust:dust:governance, dust:dust:builder. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: dust:dust:builder, dust:dust:governance, dust:dust:spaces, dust:dust:models #### LibreChat Verification status: Agent cross-checked (agent-cross-checked) Support: supported Agents combine instructions, a model, knowledge and tools, with version history and a permissioned internal catalog. Availability: generally-available Delivery: native Scope: surfaces=Customer-operated web application and server; version=Public documentation retrieved 2026-10-21; inspected source v0.8.8 / e8f3be08623663d4ad7f7241e693c94469b63bb0; plan=MIT customer-operated LibreChat; optional separately deployed services identified per answer; add_ons=none stated; deployment_mode=on-premise; region=Customer-selected infrastructure; model, OCR, search, observability and support services separately scoped; os=not specified. - create_assistants — Can users create assistants with instructions, knowledge and tools? [supported]: The Agent Builder creates reusable instructions/model/files/tools/skills configurations. Sources: librechat:librechat:agents. - creator_permissions — Can admins limit who creates them? [supported]: AGENTS CREATE and sharing permissions can be restricted by role; resource grants control access. Sources: librechat:librechat:access. - sharing_scope — Can assistants be shared with users, groups or everyone? [supported]: Agent ACLs support users, groups, roles and all authenticated users when public sharing is permitted. Sources: librechat:librechat:access. - catalogue — Is there an internal assistant directory? [supported]: Agents have a marketplace/catalog and searchable builder tool library. Sources: librechat:librechat:agents. - prompt_library — Are shared prompt templates available? [supported]: Prompt resources have use/create/share/share-public permissions and ACLs. Sources: librechat:librechat:access. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: librechat:librechat:agents, librechat:librechat:access ### AGT-02: Tools, actions and MCP Judgment rule: Record built-in tools, custom actions (OpenAPI), MCP client support (local and remote servers), authentication model per tool (shared service account or per-user OAuth), and admin approval of tools. Verify one external-tool call and its access boundary. #### Dust Verification status: Agent cross-checked (agent-cross-checked) Support: partial Dust supports built-in tools, customer MCP servers and personal or shared credentials. Availability: generally-available Delivery: native Scope: surfaces=Dust web workspace, Dust EU web workspace; version=Public source pass 2026-10-21; no logged-in deployment observed; plan=Business credit-priced plan; Enterprise or legacy differences are named per answer; add_ons=none stated; deployment_mode=multi-tenant-saas; region=US/global or EU workspace as named; EU-hosted-model restriction must be enabled separately; os=not specified. Limitations: - No external-tool call or access-boundary fixture was run. - builtin_tools — Which tools are built in? [supported]: Documented tools include knowledge search, table queries, web search, Computer, images, speech and agent orchestration. Sources: dust:dust:credits, dust:dust:computer. - custom_actions — Can admins add API actions? [supported]: Admins can connect customer-written MCP tools over secured publicly reachable URLs. Sources: dust:dust:tools. - mcp_client — Can it use remote MCP servers, and with what admin control? [supported]: Admins add remote MCP servers with automatic/static OAuth or bearer-token authentication and can restrict tools to skills. Sources: dust:dust:tools. - tool_auth — Do tools act as the user or as a shared account? [supported]: Shared credentials act as one service account; personal credentials use each user OAuth after initial admin setup. Sources: dust:dust:credentials. - tool_approval — Can admins approve or block tools, and can users confirm actions before they run? [partial]: Computer network domains can require user approval; dated connector releases describe approval for writes. A universal confirmation rule for every tool and unattended trigger was not established. Sources: dust:dust:computer, dust:dust:changelog. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: dust:dust:credits, dust:dust:computer, dust:dust:tools, dust:dust:credentials, dust:dust:changelog #### LibreChat Verification status: Agent cross-checked (agent-cross-checked) Support: partial Agent tools include file search, code, web and Actions, with local/remote MCP and configurable approval boundaries. Availability: generally-available Delivery: native Scope: surfaces=Customer-operated web application and server; version=Public documentation retrieved 2026-10-21; inspected source v0.8.8 / e8f3be08623663d4ad7f7241e693c94469b63bb0; plan=MIT customer-operated LibreChat; optional separately deployed services identified per answer; add_ons=Customer-operated or third-party MCP servers, Optional separate Code Interpreter service; deployment_mode=on-premise; region=Customer-selected infrastructure; model, OCR, search, observability and support services separately scoped; os=not specified. Limitations: - tool_approval: Agent/code environments document approval modes and individual tool-call approvals. The default policy and every external side-effect boundary were not exercised. - builtin_tools — Which tools are built in? [supported]: Agent capabilities include file search/context, web search, memory and optionally configured code execution/image tools. Sources: librechat:librechat:agents, librechat:librechat:images. - custom_actions — Can admins add API actions? [supported]: Agents support custom API Actions configured in the tool library. Sources: librechat:librechat:agents. - mcp_client — Can it use remote MCP servers, and with what admin control? [supported]: Configured MCP transports include stdio, SSE and Streamable HTTP; YAML and permissioned UI setup are documented. Sources: librechat:librechat:mcp, librechat:librechat:mcpcfg. - tool_auth — Do tools act as the user or as a shared account? [supported]: MCP supports configured credentials/headers and OAuth; credential and resource ownership scope depends on each server configuration. Sources: librechat:librechat:mcp, librechat:librechat:mcpcfg. - tool_approval — Can admins approve or block tools, and can users confirm actions before they run? [partial]: Agent/code environments document approval modes and individual tool-call approvals. The default policy and every external side-effect boundary were not exercised. Sources: librechat:librechat:agents, librechat:librechat:code. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: librechat:librechat:agents, librechat:librechat:images, librechat:librechat:mcp, librechat:librechat:mcpcfg, librechat:librechat:code ### AGT-03: Web search and deep research Judgment rule: Record web search on/off control per organisation and group, search provider, citation of web sources, deep or multi-step research, and whether queries leave the residency boundary. Admins of regulated companies often need web search off. #### Dust Verification status: Agent cross-checked (agent-cross-checked) Support: partial Agents can search the web and run multi-step Deep-dive research. Availability: generally-available Delivery: native Scope: surfaces=Dust web workspace, Dust EU web workspace; version=Public source pass 2026-10-21; no logged-in deployment observed; plan=Business credit-priced plan; Enterprise or legacy differences are named per answer; add_ons=none stated; deployment_mode=multi-tenant-saas; region=US/global or EU workspace as named; EU-hosted-model restriction must be enabled separately; os=not specified. Limitations: - Documented findings only; unestablished atomic questions remain explicit gaps. - web_search — Is web search available? [supported]: Web Search & Browse runs Google queries and reads result pages. Sources: dust:dust:web. - admin_toggle — Can admins disable it per organisation or group? [partial]: Web tools are selected in agent configuration; default Deep-dive can be disabled by admins. A universal per-group web-off rule was not established. Sources: dust:dust:web, dust:dust:deep. - search_provider — Which search provider handles queries, and where? [partial]: The guide calls the search a Google query. The intermediary provider and countries handling queries/crawling were not established. Sources: dust:dust:web. - deep_research — Is multi-step research with a report available? [supported]: Deep-dive plans, spawns parallel specialists and produces a research report. It searches Company Data and cannot use Table Query. Sources: dust:dust:deep. - web_citations — Are web sources cited? [supported]: Deep-dive promises citations to web and company sources. Sources: dust:dust:deep. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: dust:dust:web, dust:dust:deep #### LibreChat Verification status: Agent cross-checked (agent-cross-checked) Support: partial Web search combines configured search, scraping and optional reranking services; source queries can leave the installation boundary. Availability: generally-available Delivery: native Scope: surfaces=Customer-operated web application and server; version=Public documentation retrieved 2026-10-21; inspected source v0.8.8 / e8f3be08623663d4ad7f7241e693c94469b63bb0; plan=MIT customer-operated LibreChat; optional separately deployed services identified per answer; add_ons=none stated; deployment_mode=on-premise; region=Customer-selected infrastructure; model, OCR, search, observability and support services separately scoped; os=not specified. Limitations: - deep_research: Agents can perform repeated tool operations and web research. No dedicated report fixture, quality result or preset research SLA was verified. - web_search — Is web search available? [supported]: Tool-capable chats/Agents can use the configured web search capability. Sources: librechat:librechat:web, librechat:librechat:agents. - admin_toggle — Can admins disable it per organisation or group? [supported]: WEB_SEARCH feature permissions control use; operators configure whether its provider stack is available. Sources: librechat:librechat:access, librechat:librechat:web. - search_provider — Which search provider handles queries, and where? [supported]: Documented choices include Serper, SearXNG, Tavily and Keenable, with Firecrawl/Tavily/Keenable scraping and optional Jina/Cohere reranking. Geography depends on the chosen services. Sources: librechat:librechat:web. - deep_research — Is multi-step research with a report available? [partial]: Agents can perform repeated tool operations and web research. No dedicated report fixture, quality result or preset research SLA was verified. Sources: librechat:librechat:agents, librechat:librechat:web. - web_citations — Are web sources cited? [supported]: The web-search guide documents source references/citations from search and fetched content. Sources: librechat:librechat:web. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: librechat:librechat:web, librechat:librechat:agents, librechat:librechat:access ### AGT-04: Workflows and automation Judgment rule: Record no-code or low-code workflows, scheduled prompts, triggers from email, forms or business systems, and multi-step chains. Name where they run and who can build them. A shared prompt template is not a workflow. #### Dust Verification status: Agent cross-checked (agent-cross-checked) Support: partial Agents support schedules and webhook automation with admin-created event sources. Availability: generally-available Delivery: native Scope: surfaces=Dust web workspace, Dust EU web workspace; version=Public source pass 2026-10-21; no logged-in deployment observed; plan=Business credit-priced plan; Enterprise or legacy differences are named per answer; add_ons=none stated; deployment_mode=multi-tenant-saas; region=US/global or EU workspace as named; EU-hosted-model restriction must be enabled separately; os=not specified. Limitations: - Documented findings only; unestablished atomic questions remain explicit gaps. - workflow_builder — Is there a workflow or chain builder? [partial]: Agents orchestrate tools and sub-agents; external n8n and trigger workflows are documented. A separate visual multi-step chain canvas was not established. Sources: dust:dust:deep, dust:dust:triggers, dust:dust:changelog. - scheduled_tasks — Can prompts or assistants run on a schedule? [supported]: Agents can run on recurring schedules. Sources: dust:dust:schedules. - event_triggers — Can external events start a workflow? [supported]: Built-in webhook sources include GitHub, Jira, Linear, Zendesk and Fathom; signed custom webhooks are supported. Sources: dust:dust:webhooks. - builder_audience — Is building aimed at business users, IT or developers? [supported]: Admins connect event sources; builders/members attach filtered triggers in Agent Builder and can use natural-language trigger creation through Dust Default. Sources: dust:dust:webhooks, dust:dust:changelog. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: dust:dust:deep, dust:dust:triggers, dust:dust:changelog, dust:dust:schedules, dust:dust:webhooks #### LibreChat Verification status: Agent cross-checked (agent-cross-checked) Support: partial Experimental Scheduled Chats run saved Agents on cron schedules; beta Agent Events can receive external deliveries. Availability: beta Delivery: native Scope: surfaces=Customer-operated web application and server; version=Public documentation retrieved 2026-10-21; inspected source v0.8.8 / e8f3be08623663d4ad7f7241e693c94469b63bb0; plan=MIT customer-operated LibreChat; optional separately deployed services identified per answer; add_ons=Experimental schedules explicitly enabled, Beta Agents API for external Agent Events; deployment_mode=on-premise; region=Customer-selected infrastructure; model, OCR, search, observability and support services separately scoped; os=not specified. Limitations: - workflow_builder: Agent Builder supports tool workflows and subagent configurations; no general business no-code flow editor was established. - event_triggers: The beta Agents API includes authenticated Agent Events. A built-in email/form trigger catalog was not verified. - workflow_builder — Is there a workflow or chain builder? [partial]: Agent Builder supports tool workflows and subagent configurations; no general business no-code flow editor was established. Sources: librechat:librechat:agents. - scheduled_tasks — Can prompts or assistants run on a schedule? [supported]: Scheduled Chats supports hourly/daily/weekly/custom cron with time zones, owner scope and execution checks; it is experimental and off by default. Sources: librechat:librechat:schedule. - event_triggers — Can external events start a workflow? [partial]: The beta Agents API includes authenticated Agent Events. A built-in email/form trigger catalog was not verified. Sources: librechat:librechat:api. - builder_audience — Is building aimed at business users, IT or developers? [supported]: Agent Builder is a user UI; API-event and runtime configuration require technical deployment work. Sources: librechat:librechat:agents, librechat:librechat:schedule, librechat:librechat:api. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: librechat:librechat:agents, librechat:librechat:schedule, librechat:librechat:api ### AGT-05: Agentic task execution Judgment rule: Record an agent mode that plans and carries out multi-step tasks with tools (browser, files, connectors), with visible steps, user confirmation of side effects and stop control. Run one fixture task (gather three documents, draft a summary, propose an email). Record interventions and what ran. A final claim of success is not proof. #### Dust Verification status: Agent cross-checked (agent-cross-checked) Support: partial Multi-step agent work and visible Deep-dive progress are documented; safety controls depend on tool and trigger. Availability: generally-available Delivery: native Scope: surfaces=Dust web workspace, Dust EU web workspace; version=Public source pass 2026-10-21; no logged-in deployment observed; plan=Business credit-priced plan; Enterprise or legacy differences are named per answer; add_ons=none stated; deployment_mode=multi-tenant-saas; region=US/global or EU workspace as named; EU-hosted-model restriction must be enabled separately; os=not specified. Limitations: - No shared gather/documents/email fixture or side-effect audit was run. - agent_mode — Is there an agent mode for multi-step tasks? [supported]: Deep-dive plans, delegates and synthesizes, and Computer can execute deterministic file/code work. Sources: dust:dust:deep, dust:dust:computer. - visible_steps — Are steps and tool calls shown? [supported]: Deep-dive documentation shows reasoning and sub-agent progress in real time. Sources: dust:dust:deep. - side_effect_confirmation — Must the user confirm actions with side effects? [partial]: Computer domains may require user approval and selected connector write actions ask for approval. Pods agents can also act without a human, so universal write confirmation is not claimed. Sources: dust:dust:computer, dust:dust:changelog, dust:dust:pods. - stop_control — Can the user stop a running task? [partial]: The 600-AWU root-task checkpoint offers Stop or Continue. A complete general stop/rollback behavior was not tested. Sources: dust:dust:credits. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: dust:dust:deep, dust:dust:computer, dust:dust:changelog, dust:dust:pods, dust:dust:credits #### LibreChat Verification status: Agent cross-checked (agent-cross-checked) Support: partial Agent runs expose reasoning/tool activity and optional approval, interruption and cooperative stop controls. Availability: generally-available Delivery: native Scope: surfaces=Customer-operated web application and server; version=Public documentation retrieved 2026-10-21; inspected source v0.8.8 / e8f3be08623663d4ad7f7241e693c94469b63bb0; plan=MIT customer-operated LibreChat; optional separately deployed services identified per answer; add_ons=none stated; deployment_mode=on-premise; region=Customer-selected infrastructure; model, OCR, search, observability and support services separately scoped; os=not specified. Limitations: - No gather-documents, draft-summary and propose-email task was run. - Highly experimental attached workspaces are distinct from ordinary Agent tool use. - agent_mode — Is there an agent mode for multi-step tasks? [supported]: Agents execute multi-step tool workflows; code/workspace capabilities require a compatible separately deployed service. Sources: librechat:librechat:agents, librechat:librechat:code. - visible_steps — Are steps and tool calls shown? [supported]: Activity groups and tool cards distinguish preparation, execution, completion, cancellation and failure. Sources: librechat:librechat:agents. - side_effect_confirmation — Must the user confirm actions with side effects? [partial]: Attached environments can expose Ask before changes/Accept edits/Full access, bounded by operator tool rules. This is configurable, not a guarantee that all side effects require confirmation. Sources: librechat:librechat:agents, librechat:librechat:code. - stop_control — Can the user stop a running task? [partial]: The release documents interruption/steering; ordinary background-tool cancellation is optional and cooperative, so an uncooperative tool may still finish. Sources: librechat:librechat:agents, librechat:librechat:releases. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: librechat:librechat:agents, librechat:librechat:code, librechat:librechat:releases ### AGT-06: Programmable access Judgment rule: Record a public API for chat, assistants and knowledge, SDKs, API authentication, rate limits and whether API use counts against the same governance (audit, budgets, model restrictions). Being an MCP client is different from exposing the product over MCP. #### Dust Verification status: Agent cross-checked (agent-cross-checked) Support: supported Dust offers conversation/knowledge APIs, enforceable API-key caps and OAuth MCP access. Availability: generally-available Delivery: native Scope: surfaces=Dust APIs and MCP server; version=Public source pass 2026-10-21; no logged-in deployment observed; plan=Business conversation API; Enterprise adds Data Source API per pricing; MCP subject to workspace enablement; add_ons=none stated; deployment_mode=multi-tenant-saas; region=US/global or EU workspace as named; EU-hosted-model restriction must be enabled separately; os=not specified. - chat_api — Is there an API for chat and assistants? [supported]: Documented APIs retrieve/create conversations and messages with scoped API keys. Sources: dust:dust:get-conversation, dust:dust:api-keys. - knowledge_api — Can documents be loaded through an API? [supported]: The data-source API upserts documents into scoped data sources. Sources: dust:dust:upsert. - api_governance — Do audit, budgets and model rules apply to API use? [supported]: Per-key caps reject further API usage at the limit. Model overrides obey workspace/tier rules; programmatic usage uses the workspace pool and does not count toward beta pooled group budgets. Sources: dust:dust:api-keys, dust:dust:models, dust:dust:credit-management. - mcp_server — Does the product expose itself to other agents over MCP? [supported]: Global and EU MCP URLs expose authorized Dust knowledge, conversations, agents, files and Pods via OAuth; admins can disable access and control redirects. Sources: dust:dust:mcp-server. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: dust:dust:get-conversation, dust:dust:api-keys, dust:dust:upsert, dust:dust:models, dust:dust:credit-management, dust:dust:mcp-server #### LibreChat Verification status: Agent cross-checked (agent-cross-checked) Support: partial The beta Agents API provides inference, Agent management and authenticated events; deployment guards and limitations are explicit. Availability: beta Delivery: native Scope: surfaces=Customer-operated web application and server; version=Public documentation retrieved 2026-10-21; inspected source v0.8.8 / e8f3be08623663d4ad7f7241e693c94469b63bb0; plan=MIT customer-operated LibreChat; optional separately deployed services identified per answer; add_ons=none stated; deployment_mode=on-premise; region=Customer-selected infrastructure; model, OCR, search, observability and support services separately scoped; os=not specified. Limitations: - knowledge_api: The beta management surface includes Agent/Skill management. A complete public knowledge-ingestion contract and fixture were not verified. - api_governance: Agent access and remote-agent permissions gate APIs, with source identity bound for events. Complete logging/budget/model parity across routes was not tested. - mcp_server: Does the product expose itself to other agents over MCP? A native product-wide MCP server was not established; being an MCP client does not establish that interface. - chat_api — Is there an API for chat and assistants? [supported]: Chat Completions and Open Responses interfaces expose saved Agents, with API-key or configured OIDC authentication. Sources: librechat:librechat:api. - knowledge_api — Can documents be loaded through an API? [partial]: The beta management surface includes Agent/Skill management. A complete public knowledge-ingestion contract and fixture were not verified. Sources: librechat:librechat:api. - api_governance — Do audit, budgets and model rules apply to API use? [partial]: Agent access and remote-agent permissions gate APIs, with source identity bound for events. Complete logging/budget/model parity across routes was not tested. Sources: librechat:librechat:api, librechat:librechat:access. - mcp_server — Does the product expose itself to other agents over MCP? [unknown]: mcp_server: Does the product expose itself to other agents over MCP? A native product-wide MCP server was not established; being an MCP client does not establish that interface. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: librechat:librechat:api, librechat:librechat:access ## Daily work ### WORK-01: Clients and access points Judgment rule: List web, desktop (Windows, macOS), iOS, Android, mobile browser and browser extension separately. For each, record login with company SSO, feature gaps against web, and MDM/Intune deployability. An installable app is different from a mobile website. #### Dust Verification status: Agent cross-checked (agent-cross-checked) Support: partial Web, browser extension and Teams access are documented; native desktop/mobile availability remains unverified. Availability: generally-available Delivery: native Scope: surfaces=Dust web workspace, Dust EU web workspace, Dust browser extension; version=Public source pass 2026-10-21; no logged-in deployment observed; plan=Business credit-priced plan; Enterprise or legacy differences are named per answer; add_ons=none stated; deployment_mode=multi-tenant-saas; region=US/global or EU workspace as named; EU-hosted-model restriction must be enabled separately; os=not specified. Limitations: - Documented findings only; unestablished atomic questions remain explicit gaps. - web — Is a full web client available? [supported]: Dust provides the hosted web workspace, including the EU region. Sources: dust:dust:pricing, dust:dust:models. - desktop_apps — Windows and macOS apps? [unknown]: Windows and macOS apps? The documentation index, extension page, pricing and integration guides were checked. Native Windows/macOS apps, iOS/Android SSO apps, managed MDM configuration and a mobile parity matrix were not established. - mobile_apps — iOS and Android apps with SSO? [unknown]: iOS and Android apps with SSO? The documentation index, extension page, pricing and integration guides were checked. Native Windows/macOS apps, iOS/Android SSO apps, managed MDM configuration and a mobile parity matrix were not established. - browser_extension — Is there a browser extension or sidebar? [supported]: The extension guide covers Chrome, Firefox and Chromium browsers. Sources: dust:dust:browser. - managed_deployment — Can IT deploy and configure clients through MDM? [unknown]: Can IT deploy and configure clients through MDM? The documentation index, extension page, pricing and integration guides were checked. Native Windows/macOS apps, iOS/Android SSO apps, managed MDM configuration and a mobile parity matrix were not established. - client_feature_gaps — Which features are missing on mobile? [unknown]: Which features are missing on mobile? The documentation index, extension page, pricing and integration guides were checked. Native Windows/macOS apps, iOS/Android SSO apps, managed MDM configuration and a mobile parity matrix were not established. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: dust:dust:pricing, dust:dust:models, dust:dust:browser #### LibreChat Verification status: Agent cross-checked (agent-cross-checked) Support: partial The main deployment is a responsive browser client. Native clients, MDM and a supported sidebar extension were not verified. Availability: generally-available Delivery: native Scope: surfaces=Customer-operated web application and server; version=Public documentation retrieved 2026-10-21; inspected source v0.8.8 / e8f3be08623663d4ad7f7241e693c94469b63bb0; plan=MIT customer-operated LibreChat; optional separately deployed services identified per answer; add_ons=none stated; deployment_mode=on-premise; region=Customer-selected infrastructure; model, OCR, search, observability and support services separately scoped; os=not specified. Limitations: - desktop_apps: Windows and macOS apps? The self-host installation and client docs did not establish native Windows/macOS apps, a first-party browser extension or Intune/MDM configuration. - mobile_apps: Docs describe mobile navigation and temporary-chat controls in the web UI. No native iOS/Android client with company SSO was established. - browser_extension: Is there a browser extension or sidebar? The self-host installation and client docs did not establish native Windows/macOS apps, a first-party browser extension or Intune/MDM configuration. - managed_deployment: Can IT deploy and configure clients through MDM? The self-host installation and client docs did not establish native Windows/macOS apps, a first-party browser extension or Intune/MDM configuration. - client_feature_gaps: Desktop/mobile layouts expose different navigation controls; no comprehensive parity matrix was tested. - web — Is a full web client available? [supported]: A Docker/npm deployment serves a browser application at the configured server URL. Sources: librechat:librechat:firstchat, librechat:librechat:local. - desktop_apps — Windows and macOS apps? [unknown]: desktop_apps: Windows and macOS apps? The self-host installation and client docs did not establish native Windows/macOS apps, a first-party browser extension or Intune/MDM configuration. - mobile_apps — iOS and Android apps with SSO? [partial]: Docs describe mobile navigation and temporary-chat controls in the web UI. No native iOS/Android client with company SSO was established. Sources: librechat:librechat:retention, librechat:librechat:insights. - browser_extension — Is there a browser extension or sidebar? [unknown]: browser_extension: Is there a browser extension or sidebar? The self-host installation and client docs did not establish native Windows/macOS apps, a first-party browser extension or Intune/MDM configuration. - managed_deployment — Can IT deploy and configure clients through MDM? [unknown]: managed_deployment: Can IT deploy and configure clients through MDM? The self-host installation and client docs did not establish native Windows/macOS apps, a first-party browser extension or Intune/MDM configuration. - client_feature_gaps — Which features are missing on mobile? [partial]: Desktop/mobile layouts expose different navigation controls; no comprehensive parity matrix was tested. Sources: librechat:librechat:retention, librechat:librechat:insights. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: librechat:librechat:firstchat, librechat:librechat:local, librechat:librechat:retention, librechat:librechat:insights ### WORK-02: Office suite and Teams integration Judgment rule: Record presence inside Word, Excel, PowerPoint, Outlook and Teams (and Google Docs, Gmail), with what the assistant can do in each: read the open document, write into it, act on mail. Record extra licences and admin consent needed. A connector that searches Office files is KNOW-03, not WORK-02. #### Dust Verification status: Agent cross-checked (agent-cross-checked) Support: partial An official Teams integration runs agents in Teams; document connectors do not imply Office add-ins. Availability: generally-available Delivery: official-integration Scope: surfaces=Dust in Microsoft Teams; version=Public source pass 2026-10-21; no logged-in deployment observed; plan=Business credit-priced plan; Enterprise or legacy differences are named per answer; add_ons=none stated; deployment_mode=multi-tenant-saas; region=US/global or EU workspace as named; EU-hosted-model restriction must be enabled separately; os=not specified. Limitations: - Documented findings only; unestablished atomic questions remain explicit gaps. - word_excel_powerpoint — Does it work inside Word, Excel and PowerPoint? [unknown]: Does it work inside Word, Excel and PowerPoint? Teams and Microsoft connector guides were checked. Embedded assistance inside Word/Excel/PowerPoint, Outlook, Docs/Sheets/Gmail was not established; MCP tool access is recorded separately. - outlook — Does it work inside Outlook? [unknown]: Does it work inside Outlook? Teams and Microsoft connector guides were checked. Embedded assistance inside Word/Excel/PowerPoint, Outlook, Docs/Sheets/Gmail was not established; MCP tool access is recorded separately. - teams — Is there a Teams app or bot? [supported]: Dust in Teams has an official integration guide and admin setup. Sources: dust:dust:teams. - google_workspace_apps — Does it work inside Docs, Sheets or Gmail? [unknown]: Does it work inside Docs, Sheets or Gmail? Teams and Microsoft connector guides were checked. Embedded assistance inside Word/Excel/PowerPoint, Outlook, Docs/Sheets/Gmail was not established; MCP tool access is recorded separately. - licence_prerequisites — Which licences or consents are needed? [partial]: The Teams guide requires integration/admin authorization in a Dust workspace. Extra Microsoft/Dust licence charges were not established. Sources: dust:dust:teams. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: dust:dust:teams #### LibreChat Verification status: No public information found (not-published) Support: not-published We found no public information on this. A missing finding is not evidence that the product lacks the capability. ### WORK-03: Translation and writing Judgment rule: Record document translation preserving layout, the translation engine (LLM or dedicated such as DeepL), glossaries, formality control, supported languages including Swiss German conventions (no ß), and rewriting/proofreading tools. Translate the shared German contract into French and Italian and record layout and glossary results. #### Dust Verification status: Agent cross-checked (agent-cross-checked) Support: partial Agents can perform language work, but document-translation fidelity and controlled terminology are untested. Availability: generally-available Delivery: native Scope: surfaces=Dust web workspace, Dust EU web workspace; version=Public source pass 2026-10-21; no logged-in deployment observed; plan=Business credit-priced plan; Enterprise or legacy differences are named per answer; add_ons=none stated; deployment_mode=multi-tenant-saas; region=US/global or EU workspace as named; EU-hosted-model restriction must be enabled separately; os=not specified. Limitations: - Documented findings only; unestablished atomic questions remain explicit gaps. - document_translation — Can whole documents be translated with formatting kept? [unknown]: Can whole documents be translated with formatting kept? Builder, file-processing and language documentation were checked. Whole-document translation with layout retention, a dedicated engine, company glossaries, formality/de-CH controls and DE-to-FR/IT fixture outcomes were not established. - translation_engine — Which engine translates? [unknown]: Which engine translates? Builder, file-processing and language documentation were checked. Whole-document translation with layout retention, a dedicated engine, company glossaries, formality/de-CH controls and DE-to-FR/IT fixture outcomes were not established. - glossaries — Are company glossaries supported? [unknown]: Are company glossaries supported? Builder, file-processing and language documentation were checked. Whole-document translation with layout retention, a dedicated engine, company glossaries, formality/de-CH controls and DE-to-FR/IT fixture outcomes were not established. - formality_and_variants — Can formality and regional variants (de-CH) be set? [unknown]: Can formality and regional variants (de-CH) be set? Builder, file-processing and language documentation were checked. Whole-document translation with layout retention, a dedicated engine, company glossaries, formality/de-CH controls and DE-to-FR/IT fixture outcomes were not established. - writing_tools — Are dedicated rewrite and proofreading tools available? [partial]: Configurable agent instructions and Computer document editing support writing workflows; a separate dedicated proofreading product was not established. Sources: dust:dust:builder, dust:dust:computer. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: dust:dust:builder, dust:dust:computer #### LibreChat Verification status: No public information found (not-published) Support: not-published We found no public information on this. A missing finding is not evidence that the product lacks the capability. ### WORK-04: Image generation and editing Judgment rule: Record available image models, editing (inpainting, reference images), admin control and per-image cost visibility, and licence or usage terms for generated images. #### Dust Verification status: Agent cross-checked (agent-cross-checked) Support: partial Image tools use reference images and parallel generation; exact model and output-right terms need further evidence. Availability: generally-available Delivery: native Scope: surfaces=Dust web workspace, Dust EU web workspace; version=Public source pass 2026-10-21; no logged-in deployment observed; plan=Business credit-priced plan; Enterprise or legacy differences are named per answer; add_ons=none stated; deployment_mode=multi-tenant-saas; region=US/global or EU workspace as named; EU-hosted-model restriction must be enabled separately; os=not specified. Limitations: - Documented findings only; unestablished atomic questions remain explicit gaps. - image_generation — Can users generate images, with which models? [partial]: Image Generation is a configurable agent capability with SynthID-marked output. The exact current model identifier is not named in the checked guide. Sources: dust:dust:images. - image_editing — Can users edit uploaded images? [partial]: Uploaded and prior generated images can be used as references, with up to 14 references. Precise inpainting/masking controls were not established. Sources: dust:dust:images. - admin_control — Can admins restrict image generation? [partial]: Builders enable image tools in an agent, while group governance controls who creates/publishes agents. A separate organization/group image-off switch was not established. Sources: dust:dust:images, dust:dust:governance. - usage_terms — What usage rights apply to output? [unknown]: What usage rights apply to output? The image guide and legal entry points were checked. Current model identifier, per-image tariff and contractual output usage rights were not established. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: dust:dust:images, dust:dust:governance #### LibreChat Verification status: Agent cross-checked (agent-cross-checked) Support: partial Configured Agent image tools generate and edit images; model service rights/costs remain separate. Availability: generally-available Delivery: native Scope: surfaces=Customer-operated web application and server; version=Public documentation retrieved 2026-10-21; inspected source v0.8.8 / e8f3be08623663d4ad7f7241e693c94469b63bb0; plan=MIT customer-operated LibreChat; optional separately deployed services identified per answer; add_ons=none stated; deployment_mode=on-premise; region=Customer-selected infrastructure; model, OCR, search, observability and support services separately scoped; os=not specified. Limitations: - admin_control: Operators control tool credentials/configuration and Agent creation/access permissions. A dedicated per-group image-spend switch was not established. - usage_terms: What usage rights apply to output? Generated-output rights and complete per-image cost visibility were not established as a LibreChat software entitlement. - image_generation — Can users generate images, with which models? [supported]: OpenAI Image Tools and additional documented provider tools attach to an Agent; the current guide defaults OpenAI tools to gpt-image-2.5-flare. Sources: librechat:librechat:images. - image_editing — Can users edit uploaded images? [supported]: The image guide documents editing/remixing uploaded images using image IDs and compatible tools. Sources: librechat:librechat:images. - admin_control — Can admins restrict image generation? [partial]: Operators control tool credentials/configuration and Agent creation/access permissions. A dedicated per-group image-spend switch was not established. Sources: librechat:librechat:images, librechat:librechat:access. - usage_terms — What usage rights apply to output? [unknown]: usage_terms: What usage rights apply to output? Generated-output rights and complete per-image cost visibility were not established as a LibreChat software entitlement. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: librechat:librechat:images, librechat:librechat:access ### WORK-05: Voice and meetings Judgment rule: Record dictation, read-aloud, live voice conversation, meeting transcription and summarisation, supported languages including Swiss German speech, and where audio is processed. #### Dust Verification status: Agent cross-checked (agent-cross-checked) Support: partial Speech generation and meeting-transcript integrations are documented; live voice and Swiss German recognition remain unknown. Availability: generally-available Delivery: native Scope: surfaces=Dust web workspace, Dust EU web workspace; version=Public source pass 2026-10-21; no logged-in deployment observed; plan=Business credit-priced plan; Enterprise or legacy differences are named per answer; add_ons=none stated; deployment_mode=multi-tenant-saas; region=US/global or EU workspace as named; EU-hosted-model restriction must be enabled separately; os=not specified. Limitations: - Documented findings only; unestablished atomic questions remain explicit gaps. - dictation — Can users dictate prompts? [unknown]: Can users dictate prompts? Speech, upload and meeting guides were checked. Prompt dictation, live two-way voice, Swiss German accuracy and audio-processing countries were not established. - read_aloud — Can answers be read aloud? [partial]: Speech Generator converts written text to MP3 audio with selectable voices/languages; a one-click read-aloud mode for all responses was not established. Sources: dust:dust:speech. - voice_conversation — Is a live voice mode available? [unknown]: Is a live voice mode available? Speech, upload and meeting guides were checked. Prompt dictation, live two-way voice, Swiss German accuracy and audio-processing countries were not established. - meeting_summaries — Are meetings transcribed and summarised? [supported]: Connected Granola/Fathom/Gong/Praiz transcripts can feed summaries; private Gong transcripts are excluded. This is not a native meeting recorder claim. Sources: dust:dust:meetings. - swiss_german_speech — Is Swiss German speech recognised? [unknown]: Is Swiss German speech recognised? Speech, upload and meeting guides were checked. Prompt dictation, live two-way voice, Swiss German accuracy and audio-processing countries were not established. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: dust:dust:speech, dust:dust:meetings #### LibreChat Verification status: Agent cross-checked (agent-cross-checked) Support: partial LibreChat documents STT, TTS and a hands-free conversation mode that chains them. Real-time speech-to-speech behavior, meeting outcomes and Swiss German recognition remain unverified. Availability: generally-available Delivery: native Scope: surfaces=Customer-operated web application and server; version=Current self-hosted LibreChat Speech Settings documentation retrieved 2026-10-21; no voice-latency or dialect fixture performed; plan=MIT customer-operated LibreChat; optional separately deployed services identified per answer; add_ons=Optional configured STT/TTS engines and provider credentials; self-hosted speech services need allowedAddresses; deployment_mode=on-premise; region=Customer-selected app infrastructure; audio processing depends on browser or configured external/self-hosted speech engines, with no universal region verified; os=not specified. Limitations: - Conversation mode is a chain of transcription, text generation and playback; real-time audio-to-audio behavior, interruption and latency were not verified. - Audio location depends on the chosen engine. No meeting or Swiss German fixture was run. - dictation — Can users dictate prompts? [supported]: Speech configuration provides STT for audio input. Sources: librechat:librechat:speech. - read_aloud — Can answers be read aloud? [supported]: Speech configuration provides TTS for model responses. Sources: librechat:librechat:speech. - voice_conversation — Is a live voice mode available? [partial]: speechTab.conversationMode chains STT and TTS for hands-free turns: autoSendText sends the transcript after a configured delay, and automaticPlayback reads a completed response. Microphone recording is initially started by the user. Real-time speech-to-speech, interruption behavior and end-to-end latency were not tested. Sources: librechat:librechat:speech-conversation-r2. - meeting_summaries — Are meetings transcribed and summarised? [unknown]: No meeting capture/integration, speaker diarization or meeting-summary fixture was verified in the reviewed speech documentation. - swiss_german_speech — Is Swiss German speech recognised? [unknown]: Swiss German recognition and transcript quality were not tested; a language setting does not establish dialect accuracy. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: librechat:librechat:speech, librechat:librechat:speech-conversation-r2 ### WORK-06: Data analysis and document output Judgment rule: Record code execution for data analysis, charts, and generation of downloadable Word, Excel, PowerPoint and PDF files. Analyse the fixture Excel and request a PowerPoint summary. Record where code runs and whether files are editable. #### Dust Verification status: Agent cross-checked (agent-cross-checked) Support: supported Computer executes analysis and creates real Office files; Frames provide editable interactive output. Availability: generally-available Delivery: native Scope: surfaces=Dust web workspace, Dust EU web workspace; version=Public source pass 2026-10-21; no logged-in deployment observed; plan=Business credit-priced plan; Enterprise or legacy differences are named per answer; add_ons=none stated; deployment_mode=multi-tenant-saas; region=US/global or EU workspace as named; EU-hosted-model restriction must be enabled separately; os=not specified. Limitations: - No Excel/PowerPoint fixture or artifact fidelity test. File Generation is deprecated in favor of Computer as of October 14, 2026. - data_analysis — Can it analyse spreadsheets with executed code? [supported]: Computer executes code to inspect and transform Excel/CSV, compute every-row results and return edited workbooks. Sources: dust:dust:computer. - charts — Can it produce charts? [supported]: Computer documents chart generation from spreadsheet data and inclusion in reports/decks. Sources: dust:dust:computer. - office_file_output — Can it create Word, Excel or PowerPoint files? [supported]: Computer creates/edits Word, Excel and PowerPoint files and returns downloadable artifacts, with fidelity caveats. Sources: dust:dust:computer. - canvas_editing — Is there a document or canvas editor for iterating on output? [supported]: Frames are persistent interactive content that can be revised through agents and shared. Sources: dust:dust:frames. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: dust:dust:computer, dust:dust:frames #### LibreChat Verification status: Agent cross-checked (agent-cross-checked) Support: partial The separate Code Interpreter runs analysis and produces downloadable files; artifacts provide code/HTML/diagram previews. Availability: generally-available Delivery: native Scope: surfaces=Customer-operated web application and server; version=Public documentation retrieved 2026-10-21; inspected source v0.8.8 / e8f3be08623663d4ad7f7241e693c94469b63bb0; plan=MIT customer-operated LibreChat; optional separately deployed services identified per answer; add_ons=Separately operated compatible Code Interpreter service, Experimental stateful workspaces only if explicitly configured; deployment_mode=on-premise; region=Customer-selected infrastructure; model, OCR, search, observability and support services separately scoped; os=not specified. Limitations: - charts: Code execution and generated-file downloads support chart artifacts, but the shared spreadsheet/chart fixture was not run. - office_file_output: Generated files are downloadable; editable Word/Excel/PowerPoint fidelity was not established with a fixture. - canvas_editing: Artifacts render React/HTML/Mermaid content and source; a rich Office document editor was not established. - data_analysis — Can it analyse spreadsheets with executed code? [supported]: Configured Code Interpreter executes programs against uploaded files in its separate sandbox. Sources: librechat:librechat:code. - charts — Can it produce charts? [partial]: Code execution and generated-file downloads support chart artifacts, but the shared spreadsheet/chart fixture was not run. Sources: librechat:librechat:code. - office_file_output — Can it create Word, Excel or PowerPoint files? [partial]: Generated files are downloadable; editable Word/Excel/PowerPoint fidelity was not established with a fixture. Sources: librechat:librechat:code. - canvas_editing — Is there a document or canvas editor for iterating on output? [partial]: Artifacts render React/HTML/Mermaid content and source; a rich Office document editor was not established. Sources: librechat:librechat:artifacts. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: librechat:librechat:code, librechat:librechat:artifacts ### WORK-07: Sharing and team spaces Judgment rule: Record chat sharing (link, user, group), shared projects or team spaces, channels or group chat with AI, and commenting. Test sharing a chat with a colleague in another group and record what the recipient sees, including knowledge they lack access to. #### Dust Verification status: Agent cross-checked (agent-cross-checked) Support: partial Pods provide shared team conversations, files and tasks; source-sharing effects still need testing. Availability: generally-available Delivery: native Scope: surfaces=Dust web workspace, Dust EU web workspace; version=Public source pass 2026-10-21; no logged-in deployment observed; plan=Business credit-priced plan; Enterprise or legacy differences are named per answer; add_ons=none stated; deployment_mode=multi-tenant-saas; region=US/global or EU workspace as named; EU-hosted-model restriction must be enabled separately; os=not specified. Limitations: - No cross-group sharing fixture. Do not infer source-ACL propagation from Pod membership. - share_chats — Can chats be shared inside the organisation? [supported]: Pod conversations are visible to all members, who can contribute. Sources: dust:dust:pod-chats. - team_spaces — Are shared projects or workspaces available? [supported]: Open/Restricted Pods group conversations, files and tasks; restricted access is invite-only. Sources: dust:dust:pods. - group_chat — Can several people chat together with the AI? [supported]: Teammates and agents participate in the same Pod thread and can be mentioned. Sources: dust:dust:pod-chats. - share_permission_boundary — Does sharing respect the recipient's access to sources? [partial]: All Pod members read conversation contents. Spaces restrict source/agent use and personal tools act under their user; recipient-safe redaction of previously generated text was not established. Sources: dust:dust:pod-chats, dust:dust:permissions, dust:dust:credentials. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: dust:dust:pod-chats, dust:dust:pods, dust:dust:permissions, dust:dust:credentials #### LibreChat Verification status: Agent cross-checked (agent-cross-checked) Support: partial Shared chat links publish transcript snapshots and optional files. Projects are personal, and shared source permission boundaries need care. Availability: generally-available Delivery: native Scope: surfaces=Customer-operated web application and server; version=Public documentation retrieved 2026-10-21; inspected source v0.8.8 / e8f3be08623663d4ad7f7241e693c94469b63bb0; plan=MIT customer-operated LibreChat; optional separately deployed services identified per answer; add_ons=none stated; deployment_mode=on-premise; region=Customer-selected infrastructure; model, OCR, search, observability and support services separately scoped; os=not specified. Limitations: - group_chat: Can several people chat together with the AI? Simultaneous multi-person AI group chat was not established from Projects and shareable-link documentation; link continuation creates a separate personal copy. - share_permission_boundary: Published file snapshots are served independently of the owner’s live file ACL. Sharing grants access to the published copy, not continuing source-system authorization; no restricted-source fixture was run. - share_chats — Can chats be shared inside the organisation? [supported]: Authenticated shared links are supported, with optional public viewing; public unauthenticated access is off by default. Sources: librechat:librechat:share. - team_spaces — Are shared projects or workspaces available? [unsupported]: Projects specifically cannot be shared or transferred and are personal to the creator. Shared Agents/resources are a separate capability. Sources: librechat:librechat:projects. - group_chat — Can several people chat together with the AI? [unknown]: group_chat: Can several people chat together with the AI? Simultaneous multi-person AI group chat was not established from Projects and shareable-link documentation; link continuation creates a separate personal copy. - share_permission_boundary — Does sharing respect the recipient's access to sources? [partial]: Published file snapshots are served independently of the owner’s live file ACL. Sharing grants access to the published copy, not continuing source-system authorization; no restricted-source fixture was run. Sources: librechat:librechat:share. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: librechat:librechat:share, librechat:librechat:projects ### WORK-08: Interface languages and accessibility Judgment rule: Record UI languages, focusing on German, French, Italian and English, with translation quality observed (complete, partial, machine-translated). Record response language handling and WCAG conformance statements. #### Dust Verification status: Agent cross-checked (agent-cross-checked) Support: partial German and French UI releases are documented alongside English, with remaining English areas. Availability: generally-available Delivery: native Scope: surfaces=Dust web workspace, Dust EU web workspace; version=Public source pass 2026-10-21; no logged-in deployment observed; plan=Business credit-priced plan; Enterprise or legacy differences are named per answer; add_ons=none stated; deployment_mode=multi-tenant-saas; region=US/global or EU workspace as named; EU-hosted-model restriction must be enabled separately; os=not specified. Limitations: - Documented findings only; unestablished atomic questions remain explicit gaps. - ui_languages — Which UI languages are offered? [supported]: October 21, 2026 release adds German; October 16 adds French alongside English US/UK, across product, Frames, extension, Sidekick and Slack/Teams. Sources: dust:dust:changelog. - dach_language_quality — Are German, French and Italian UIs complete and natural? [partial]: Release notes explicitly leave agent/skill names and instructions in English. German/French naturalness and completeness were not observed; Italian UI was not established. Sources: dust:dust:changelog. - accessibility_statement — Is there a WCAG conformance statement? [unknown]: Is there a WCAG conformance statement? The dated locale announcements were checked; Italian UI and a WCAG conformance statement/report were not established. - admin_ui_languages — Is the admin interface available in German? [partial]: Admins can set a workspace default language. A complete German admin-screen walkthrough was not observed. Sources: dust:dust:changelog. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: dust:dust:changelog #### LibreChat Verification status: Agent cross-checked (agent-cross-checked) Support: partial English, German, French and Italian UI resources are registered in the pinned client; quality and accessibility conformance were not evaluated. Availability: generally-available Delivery: native Scope: surfaces=Customer-operated web application and server; version=Public documentation retrieved 2026-10-21; inspected source v0.8.8 / e8f3be08623663d4ad7f7241e693c94469b63bb0; plan=MIT customer-operated LibreChat; optional separately deployed services identified per answer; add_ons=none stated; deployment_mode=on-premise; region=Customer-selected infrastructure; model, OCR, search, observability and support services separately scoped; os=not specified. Limitations: - dach_language_quality: Are German, French and Italian UIs complete and natural? No complete DE/FR/IT walkthrough, German separate-admin-panel coverage test or WCAG conformance statement/report was verified. - accessibility_statement: Is there a WCAG conformance statement? No complete DE/FR/IT walkthrough, German separate-admin-panel coverage test or WCAG conformance statement/report was verified. - admin_ui_languages: Is the admin interface available in German? No complete DE/FR/IT walkthrough, German separate-admin-panel coverage test or WCAG conformance statement/report was verified. - ui_languages — Which UI languages are offered? [supported]: The client i18n implementation registers en, de, fr and it among community-maintained translations. Sources: librechat:librechat:i18n-pin. - dach_language_quality — Are German, French and Italian UIs complete and natural? [unknown]: dach_language_quality: Are German, French and Italian UIs complete and natural? No complete DE/FR/IT walkthrough, German separate-admin-panel coverage test or WCAG conformance statement/report was verified. - accessibility_statement — Is there a WCAG conformance statement? [unknown]: accessibility_statement: Is there a WCAG conformance statement? No complete DE/FR/IT walkthrough, German separate-admin-panel coverage test or WCAG conformance statement/report was verified. - admin_ui_languages — Is the admin interface available in German? [unknown]: admin_ui_languages: Is the admin interface available in German? No complete DE/FR/IT walkthrough, German separate-admin-panel coverage test or WCAG conformance statement/report was verified. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: librechat:librechat:i18n-pin ## Administration and governance ### ADM-01: Single sign-on and authentication Judgment rule: Record SAML, OIDC, Microsoft Entra ID, Google, LDAP, MFA enforcement, conditional-access compatibility, domain verification and plan restrictions. Test one SSO login on a trial where possible. #### Dust Verification status: Agent cross-checked (agent-cross-checked) Support: partial SAML SSO can be enforced; Business offers it from five seats on demand. Availability: generally-available Delivery: native Scope: surfaces=Dust web workspace, Dust EU web workspace; version=Public source pass 2026-10-21; no logged-in deployment observed; plan=Business credit-priced plan; Enterprise or legacy differences are named per answer; add_ons=none stated; deployment_mode=multi-tenant-saas; region=US/global or EU workspace as named; EU-hosted-model restriction must be enabled separately; os=not specified. Limitations: - No self-serve SSO login test. - sso_protocols — Which SSO protocols and identity providers work? [supported]: SAML SSO is documented, with Okta, Entra ID and JumpCloud named on pricing. Enterprise marketing also names OIDC; a full provider/protocol matrix was not verified. Sources: dust:dust:sso, dust:dust:pricing, dust:dust:enterprise. - sso_plan — Which plan includes SSO? [supported]: Business pricing lists SSO for 5+ seats on demand; Enterprise includes it. Sources: dust:dust:pricing. - mfa — Can MFA be enforced? [partial]: Enforced SAML routes sign-in through the IdP. Specific MFA enforcement behavior was not tested or established in the checked setup guide. Sources: dust:dust:sso. - conditional_access — Do Entra conditional-access policies apply? [unknown]: Do Entra conditional-access policies apply? SSO and plan guides were checked; conditional-access compatibility and verified-domain capture/migration semantics were not established. - domain_capture — Can the company claim its email domain? [unknown]: Can the company claim its email domain? SSO and plan guides were checked; conditional-access compatibility and verified-domain capture/migration semantics were not established. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: dust:dust:sso, dust:dust:pricing, dust:dust:enterprise #### LibreChat Verification status: Agent cross-checked (agent-cross-checked) Support: partial Self-hosted authentication supports OAuth/OIDC, SAML and LDAP; local/LDAP TOTP can be enforced. Availability: generally-available Delivery: native Scope: surfaces=Customer-operated web application and server; version=Public documentation retrieved 2026-10-21; inspected source v0.8.8 / e8f3be08623663d4ad7f7241e693c94469b63bb0; plan=MIT customer-operated LibreChat; optional separately deployed services identified per answer; add_ons=none stated; deployment_mode=on-premise; region=Customer-selected infrastructure; model, OCR, search, observability and support services separately scoped; os=not specified. Limitations: - conditional_access: Entra authentication and token reuse are documented, but an actual conditional-access/device-policy login was not tested. - domain_capture: Can the company claim its email domain? Registration/domain filters were not established as a tenant-wide company-domain ownership claim. - sso_protocols — Which SSO protocols and identity providers work? [supported]: OIDC/OAuth providers include Entra, Google, Keycloak and Auth0; SAML and LDAP are documented separately. Sources: librechat:librechat:openid, librechat:librechat:saml, librechat:librechat:auth. - sso_plan — Which plan includes SSO? [supported]: The authentication configuration belongs to the MIT self-hosted project, with no separate SSO licence tariff. Sources: librechat:librechat:license, librechat:librechat:auth. - mfa — Can MFA be enforced? [supported]: ENFORCE_TWO_FACTOR_AUTHENTICATION requires local/LDAP enrollment and prevents disabling it; federated accounts depend on IdP MFA. Sources: librechat:librechat:auth. - conditional_access — Do Entra conditional-access policies apply? [partial]: Entra authentication and token reuse are documented, but an actual conditional-access/device-policy login was not tested. Sources: librechat:librechat:entra. - domain_capture — Can the company claim its email domain? [unknown]: domain_capture: Can the company claim its email domain? Registration/domain filters were not established as a tenant-wide company-domain ownership claim. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: librechat:librechat:openid, librechat:librechat:saml, librechat:librechat:auth, librechat:librechat:license, librechat:librechat:entra ### ADM-02: User provisioning and lifecycle Judgment rule: Record SCIM, just-in-time provisioning, group sync from the identity provider, deprovisioning effects (chats, assistants, knowledge ownership transfer) and licence assignment. Deprovision a test user and record what happens to their content. #### Dust Verification status: Agent cross-checked (agent-cross-checked) Support: partial Enterprise SCIM and group roles are documented, with specific annual-seat reclaim behavior. Availability: generally-available Delivery: native Scope: surfaces=Dust web workspace, Dust EU web workspace; version=Public source pass 2026-10-21; no logged-in deployment observed; plan=Enterprise for SCIM; seat-management differences stated; add_ons=none stated; deployment_mode=multi-tenant-saas; region=US/global or EU workspace as named; EU-hosted-model restriction must be enabled separately; os=not specified. Limitations: - Documented findings only; unestablished atomic questions remain explicit gaps. - scim — Is SCIM provisioning supported? [supported]: SCIM 2.0 creates, updates and deactivates users from an IdP as an Enterprise feature. Sources: dust:dust:provision. - group_sync — Are identity-provider groups synced? [supported]: Provisioned IdP groups govern Space access and can map to Admin/Manager roles. Sources: dust:dust:provision, dust:dust:spaces. - jit_provisioning — Are users created at first login? [unknown]: Are users created at first login? Provisioning and seat docs were checked. First-login JIT creation and complete leaver-data transfer were not established. - deprovisioning — What happens to a removed user's data and shared assets? [partial]: SCIM deactivates users; removal/reassignment affects seats but exact chat/agent/knowledge ownership transfer was not tested. Sources: dust:dust:provision, dust:dust:seats. - seat_assignment — Can admins assign and reclaim seats? [supported]: Admins assign seat types independent of roles. Unassigned members cannot interact with agents; annual seats return to stock, and already-granted credits can persist to period end. Sources: dust:dust:seats. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: dust:dust:provision, dust:dust:spaces, dust:dust:seats #### LibreChat Verification status: Agent cross-checked (agent-cross-checked) Support: partial Entra group synchronization and federated account creation are documented; SCIM and leaver content transfer remain unknown. Availability: generally-available Delivery: native Scope: surfaces=Customer-operated web application and server; version=Public documentation retrieved 2026-10-21; inspected source v0.8.8 / e8f3be08623663d4ad7f7241e693c94469b63bb0; plan=MIT customer-operated LibreChat; optional separately deployed services identified per answer; add_ons=none stated; deployment_mode=on-premise; region=Customer-selected infrastructure; model, OCR, search, observability and support services separately scoped; os=not specified. Limitations: - scim: Is SCIM provisioning supported? SCIM was not established in the authentication/access docs or pinned repository path review. No two-user deprovisioning fixture was run. - deprovisioning: The docs provide user deletion scripts, but content ownership transfer and revocation behavior were not fixture-tested. - seat_assignment: Operators manage accounts/roles under an MIT licence; there is no per-seat licence entitlement in the scoped source. - scim — Is SCIM provisioning supported? [unknown]: scim: Is SCIM provisioning supported? SCIM was not established in the authentication/access docs or pinned repository path review. No two-user deprovisioning fixture was run. - group_sync — Are identity-provider groups synced? [supported]: With Entra OIDC token reuse, Graph groups sync at login; resource ACL principal membership has a documented cache. Sources: librechat:librechat:access, librechat:librechat:entra. - jit_provisioning — Are users created at first login? [supported]: Federated login creates accounts according to the authentication configuration; OpenID/SAML registration cannot be independently disabled. Sources: librechat:librechat:auth. - deprovisioning — What happens to a removed user's data and shared assets? [partial]: The docs provide user deletion scripts, but content ownership transfer and revocation behavior were not fixture-tested. Sources: librechat:librechat:auth. - seat_assignment — Can admins assign and reclaim seats? [partial]: Operators manage accounts/roles under an MIT licence; there is no per-seat licence entitlement in the scoped source. Sources: librechat:librechat:auth, librechat:librechat:license. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: librechat:librechat:access, librechat:librechat:entra, librechat:librechat:auth, librechat:librechat:license ### ADM-03: Roles, groups and feature permissions Judgment rule: Record admin roles, delegated admins, groups, per-feature permissions (web search, image generation, connectors, file upload, agents), separation of subsidiaries or departments, and enforcement. Test one disabled feature for one group. #### Dust Verification status: Agent cross-checked (agent-cross-checked) Support: partial Dust separates Admin, Manager, Member and group-manager permissions and supports restricted knowledge spaces. Availability: generally-available Delivery: native Scope: surfaces=Dust web workspace, Dust EU web workspace; version=Public source pass 2026-10-21; no logged-in deployment observed; plan=Business credit-priced plan; Enterprise or legacy differences are named per answer; add_ons=none stated; deployment_mode=multi-tenant-saas; region=US/global or EU workspace as named; EU-hosted-model restriction must be enabled separately; os=not specified. Limitations: - Documented findings only; unestablished atomic questions remain explicit gaps. - admin_roles — Which admin roles exist, and can admin rights be delegated? [supported]: Admins control all governance; Managers delegate member/group/usage administration; group managers administer selected groups. Sources: dust:dust:governance. - groups — Can permissions be set per group? [supported]: Permissions go to manual or provisioned groups using Everyone, Groups or Admin-only modes. Multiple grants combine additively. Sources: dust:dust:governance. - feature_toggles — Which features can be switched on or off per group? [partial]: Group controls cover agent/skill creation and publishing, automation charging, billing/security and model ceilings. An exhaustive per-feature toggle matrix for uploads/web/images was not established. Sources: dust:dust:governance, dust:dust:tiers. - multi_entity — Can subsidiaries or departments be separated within one contract? [partial]: Restricted Spaces and Pods separate audiences inside a workspace; separate legal/subsidiary tenant boundaries within one contract were not established. Sources: dust:dust:spaces, dust:dust:pods. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: dust:dust:governance, dust:dust:tiers, dust:dust:spaces, dust:dust:pods #### LibreChat Verification status: Agent cross-checked (agent-cross-checked) Support: partial Role permissions, resource ACLs and system grants support granular governance; the separate management UI remains preview. Availability: generally-available Delivery: native Scope: surfaces=Customer-operated web application and server, Separately deployed LibreChat Admin Panel preview; version=Public documentation retrieved 2026-10-21; inspected source v0.8.8 / e8f3be08623663d4ad7f7241e693c94469b63bb0; plan=MIT customer-operated LibreChat; optional separately deployed services identified per answer; add_ons=Separate Admin Panel preview for GUI management; deployment_mode=on-premise; region=Customer-selected infrastructure; model, OCR, search, observability and support services separately scoped; os=not specified. Limitations: - Separate admin management UI is preview; the core authorization model is documented as available. - No two-group disabled-feature fixture was run. - admin_roles — Which admin roles exist, and can admin rights be delegated? [supported]: ADMIN/USER and custom roles exist; capability grants can delegate selected administrative functions without full admin. Sources: librechat:librechat:access. - groups — Can permissions be set per group? [supported]: Local and Entra groups can receive ACLs, system grants and configuration overrides. Sources: librechat:librechat:access, librechat:librechat:admin. - feature_toggles — Which features can be switched on or off per group? [supported]: Permissions include Agents, MCP, prompts, memory, search, file search/citations and code execution. Sources: librechat:librechat:access. - multi_entity — Can subsidiaries or departments be separated within one contract? [partial]: Tenant-scoped API queries and group/configuration boundaries are documented. A commercial multi-entity contract and tested tenant isolation were not established. Sources: librechat:librechat:insights, librechat:librechat:access. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: librechat:librechat:access, librechat:librechat:admin, librechat:librechat:insights ### ADM-04: Usage and adoption analytics Judgment rule: Record dashboards for active users, usage by feature, model, group and assistant, export, and privacy controls (aggregation, pseudonymisation, works-council-friendly settings). Distinguish aggregate adoption metrics from per-person monitoring. #### Dust Verification status: Agent cross-checked (agent-cross-checked) Support: partial Consumption analytics has detailed attribution and exports without message content. Availability: generally-available Delivery: native Scope: surfaces=Dust web workspace, Dust EU web workspace; version=Public source pass 2026-10-21; no logged-in deployment observed; plan=Business credit-priced plan; Enterprise or legacy differences are named per answer; add_ons=none stated; deployment_mode=multi-tenant-saas; region=US/global or EU workspace as named; EU-hosted-model restriction must be enabled separately; os=not specified. Limitations: - Documented findings only; unestablished atomic questions remain explicit gaps. - adoption_dashboard — Is there an adoption dashboard? [partial]: Pricing lists usage/adoption reporting; the Analytics guide focuses on credit consumption and attribution rather than a complete active-user adoption scorecard. Sources: dust:dust:pricing, dust:dust:analytics. - breakdowns — Can usage be broken down by group, model, feature and assistant? [supported]: Analytics filters/ranks agents, members, groups, models, tools, sources, API keys, skills and Pods. Group attribution may count a member in several groups. Sources: dust:dust:analytics. - export — Can usage data be exported? [supported]: Raw CSV consumption export and API analytics export are documented, with workspace enablement and key-scope restrictions. Sources: dust:dust:analytics. - privacy_modes — Can per-person reporting be limited or pseudonymised? [partial]: Analytics exports omit message content, but contain member/message/conversation identifiers. Pseudonymous or works-council privacy modes were not established. Sources: dust:dust:analytics. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: dust:dust:pricing, dust:dust:analytics #### LibreChat Verification status: Agent cross-checked (agent-cross-checked) Support: partial Optional Insights reports identifiable activity; token transactions and optional tracing are separate sources of usage detail. Availability: generally-available Delivery: native Scope: surfaces=Customer-operated web application and server; version=Public documentation retrieved 2026-10-21; inspected source v0.8.8 / e8f3be08623663d4ad7f7241e693c94469b63bb0; plan=MIT customer-operated LibreChat; optional separately deployed services identified per answer; add_ons=none stated; deployment_mode=on-premise; region=Customer-selected infrastructure; model, OCR, search, observability and support services separately scoped; os=not specified. Limitations: - breakdowns: Insights supports Agent filtering and user activity; recorded token/context cost details are per model/endpoint. Complete group/feature/assistant exports were not established. - export: Can usage data be exported? A documented complete admin analytics export and group/cost-centre dataset were not verified. - privacy_modes: Insights is off unless enabled and is capability-gated; it can display names, emails and first-message content. Pseudonymized reporting was not established. - adoption_dashboard — Is there an adoption dashboard? [supported]: Opt-in Insights reports conversations, unique users, messages, tokens, top users and recent conversations. Sources: librechat:librechat:insights. - breakdowns — Can usage be broken down by group, model, feature and assistant? [partial]: Insights supports Agent filtering and user activity; recorded token/context cost details are per model/endpoint. Complete group/feature/assistant exports were not established. Sources: librechat:librechat:insights, librechat:librechat:tokens. - export — Can usage data be exported? [unknown]: export: Can usage data be exported? A documented complete admin analytics export and group/cost-centre dataset were not verified. - privacy_modes — Can per-person reporting be limited or pseudonymised? [partial]: Insights is off unless enabled and is capability-gated; it can display names, emails and first-message content. Pseudonymized reporting was not established. Sources: librechat:librechat:insights. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: librechat:librechat:insights, librechat:librechat:tokens ### ADM-05: Spend controls and cost allocation Judgment rule: Record cost visibility per user, group and cost centre, budgets and their behaviour (alert, block, downgrade model), pooled credits, and chargeback exports. Test a small budget on a trial where possible. Distinguish billed amounts, estimates and unavailable amounts. #### Dust Verification status: Agent cross-checked (agent-cross-checked) Support: partial Credit plans offer consumption visibility, member limits and plan-specific pooled budgets, with programmatic exceptions. Availability: generally-available Delivery: native Scope: surfaces=Dust web workspace, Dust EU web workspace; version=Public source pass 2026-10-21; no logged-in deployment observed; plan=Business credit-priced plan; Enterprise or legacy differences are named per answer; add_ons=none stated; deployment_mode=multi-tenant-saas; region=US/global or EU workspace as named; EU-hosted-model restriction must be enabled separately; os=not specified. Limitations: - Group budgets are beta and exclude programmatic usage. No budget fixture test. - cost_visibility — Can admins see cost or consumption per user and group? [supported]: Analytics and Credits show consumed credits by member, group, model, tool and agent. Credits are not assumed to be invoice currency. Sources: dust:dust:analytics, dust:dust:credit-management. - budgets — Can budgets be set per user or group? [supported]: Members have shared-pool spend limits; beta group budgets require Enterprise Pooled. Programmatic monthly and API-key caps are separate. Sources: dust:dust:credit-management, dust:dust:api-keys. - budget_enforcement — Does reaching a budget alert, block or downgrade? [supported]: Pool limits/group budgets block further requests; an in-progress message can finish above its starting cap. API calls reject at per-key caps. Optional automatic seat upgrades may increase billed subscriptions. Sources: dust:dust:credit-management, dust:dust:api-keys. - chargeback — Can costs be exported per cost centre? [partial]: Consumption CSV includes groups and identifiers, but group attribution can duplicate usage and is not a guaranteed invoice-ready cost-center chargeback. Sources: dust:dust:analytics. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: dust:dust:analytics, dust:dust:credit-management, dust:dust:api-keys #### LibreChat Verification status: Agent cross-checked (agent-cross-checked) Support: partial Optional token-credit balances block admission when insufficient; credit estimates differ from actual cloud invoices. Availability: generally-available Delivery: native Scope: surfaces=Customer-operated web application and server; version=Public documentation retrieved 2026-10-21; inspected source v0.8.8 / e8f3be08623663d4ad7f7241e693c94469b63bb0; plan=MIT customer-operated LibreChat; optional separately deployed services identified per answer; add_ons=none stated; deployment_mode=on-premise; region=Customer-selected infrastructure; model, OCR, search, observability and support services separately scoped; os=not specified. Limitations: - cost_visibility: Token transactions and configured token prices support per-user/model estimates. Insights explicitly reports activity rather than monetary billing. - budgets: Optional per-user start balances and automatic refills are documented. Group/cost-centre pooled budgets were not established. - chargeback: Can costs be exported per cost centre? Cost-centre chargeback exports and reconciliation to actual vendor invoices were not established. - cost_visibility — Can admins see cost or consumption per user and group? [partial]: Token transactions and configured token prices support per-user/model estimates. Insights explicitly reports activity rather than monetary billing. Sources: librechat:librechat:tokens, librechat:librechat:insights. - budgets — Can budgets be set per user or group? [partial]: Optional per-user start balances and automatic refills are documented. Group/cost-centre pooled budgets were not established. Sources: librechat:librechat:balance. - budget_enforcement — Does reaching a budget alert, block or downgrade? [supported]: Pinned balance middleware reserves estimated token cost and throws an error if credits cannot be reserved; no low-budget trial was run. Sources: librechat:librechat:balance-pin. - chargeback — Can costs be exported per cost centre? [unknown]: chargeback: Can costs be exported per cost centre? Cost-centre chargeback exports and reconciliation to actual vendor invoices were not established. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: librechat:librechat:tokens, librechat:librechat:insights, librechat:librechat:balance, librechat:librechat:balance-pin ### ADM-06: Audit logs and compliance export Judgment rule: Record which events are logged (logins, admin changes, prompts, responses, file access, connector queries), retention, export or SIEM streaming, and eDiscovery/legal hold. Use the word audit only with a defined retention and integrity model. #### Dust Verification status: Agent cross-checked (agent-cross-checked) Support: partial Enterprise has documented event logs and SIEM delivery; content logging and legal hold remain unverified. Availability: generally-available Delivery: native Scope: surfaces=Dust web workspace, Dust EU web workspace, Dust APIs and MCP server; version=Public source pass 2026-10-21; no logged-in deployment observed; plan=Enterprise; add_ons=none stated; deployment_mode=multi-tenant-saas; region=US/global or EU workspace as named; EU-hosted-model restriction must be enabled separately; os=not specified. Limitations: - Documented findings only; unestablished atomic questions remain explicit gaps. - admin_events — Are admin and security events logged? [supported]: Enterprise Audit Logs record significant admin/security, user, API-key, system and tool events with timestamps, targets and actor metadata. Sources: dust:dust:audit. - content_logging — Can prompts and answers be retained for compliance, and is it configurable? [unknown]: Can prompts and answers be retained for compliance, and is it configurable? Audit, analytics and public legal entry points were checked. Prompt/response compliance capture, eDiscovery/legal hold and detailed log-integrity implementation/retention contract were not established. - siem_export — Can logs stream to a SIEM or be exported? [supported]: CSV export and continuous destinations include Datadog, Splunk, S3, GCS and custom HTTPS. Sources: dust:dust:audit. - ediscovery — Are eDiscovery and legal hold supported? [unknown]: Are eDiscovery and legal hold supported? Audit, analytics and public legal entry points were checked. Prompt/response compliance capture, eDiscovery/legal hold and detailed log-integrity implementation/retention contract were not established. - log_retention — How long are logs kept? [partial]: Enterprise marketing states 365-day audit retention; no underlying report/contract or configurable retention details were retrieved. Sources: dust:dust:enterprise. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: dust:dust:audit, dust:dust:enterprise #### LibreChat Verification status: Agent cross-checked (agent-cross-checked) Support: partial Operational logs and persisted chats are available; a complete audit retention/integrity or eDiscovery system was not verified. Availability: generally-available Delivery: native Scope: surfaces=Customer-operated web application and server; version=Public documentation retrieved 2026-10-21; inspected source v0.8.8 / e8f3be08623663d4ad7f7241e693c94469b63bb0; plan=MIT customer-operated LibreChat; optional separately deployed services identified per answer; add_ons=none stated; deployment_mode=on-premise; region=Customer-selected infrastructure; model, OCR, search, observability and support services separately scoped; os=not specified. Limitations: - admin_events: Backend logs capture errors/debug activity and can include operational/security actions. A complete immutable admin-event schema was not established. - content_logging: Chats are persisted in MongoDB subject to retention settings; external tracing is optional. This is not a verified compliance archive. - siem_export: Operators can feed log files to external systems such as ELK. A native managed SIEM stream and integrity guarantee were not verified. - ediscovery: Are eDiscovery and legal hold supported? No eDiscovery/legal hold, tamper-evident archive or complete prompt/file/connector event coverage was verified. - log_retention: The logging guide describes 14-day rotation. Rotation is not evidence of an enforced legal retention or deletion guarantee across external sinks. - admin_events — Are admin and security events logged? [partial]: Backend logs capture errors/debug activity and can include operational/security actions. A complete immutable admin-event schema was not established. Sources: librechat:librechat:logging. - content_logging — Can prompts and answers be retained for compliance, and is it configurable? [partial]: Chats are persisted in MongoDB subject to retention settings; external tracing is optional. This is not a verified compliance archive. Sources: librechat:librechat:retention, librechat:librechat:compose-pin. - siem_export — Can logs stream to a SIEM or be exported? [partial]: Operators can feed log files to external systems such as ELK. A native managed SIEM stream and integrity guarantee were not verified. Sources: librechat:librechat:logging. - ediscovery — Are eDiscovery and legal hold supported? [unknown]: ediscovery: Are eDiscovery and legal hold supported? No eDiscovery/legal hold, tamper-evident archive or complete prompt/file/connector event coverage was verified. - log_retention — How long are logs kept? [partial]: The logging guide describes 14-day rotation. Rotation is not evidence of an enforced legal retention or deletion guarantee across external sinks. Sources: librechat:librechat:logging. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: librechat:librechat:logging, librechat:librechat:retention, librechat:librechat:compose-pin ### ADM-07: Data loss prevention and personal data Judgment rule: Record detection and handling of personal and confidential data before it reaches a model: redaction, masking, blocking, warning. Record integration with Purview, Netskope or similar, and whether filtering runs locally or at a third party. Send a fixture prompt with a Swiss AHV number and IBAN and record the outcome. #### Dust Verification status: No public information found (not-published) Support: not-published We found no public information on this. A missing finding is not evidence that the product lacks the capability. #### LibreChat Verification status: Agent cross-checked (agent-cross-checked) Support: partial Optional server-side content filters support configured pattern detection, blocking and auditing, plus narrowly scoped Agent text redaction. External DLP integration and fixture accuracy remain unknown. Availability: source-only Delivery: native Scope: surfaces=Customer-operated web application and server; version=Development source a378657583567310c841d652b0f6b00c83e2bd6b; current Latest (dev) documentation; limited redaction announced in config v1.3.18 on 2026-10-22; plan=Customer-operated LibreChat; optional base-YAML content policies, disabled unless configured; add_ons=none stated; deployment_mode=on-premise; region=Customer-selected infrastructure; model, OCR, search, observability and support services separately scoped; os=not specified. Limitations: - Development-source finding; no running instance or AHV/IBAN fixture. - Filtering is opt-in and limited to configured sources/fields. Default patterns cover credential shapes, not every PII class. - Redaction applies only to eligible fresh Agent text; encrypted originals remain stored. - A dedicated external DLP integration remains unverified. - pii_detection — Is personal data detected in prompts and files? [partial]: Configured source-aware policies can inspect selected prompt fields and file names/extracted text/transcripts. Built-in starter rules match credentials; admins must supply additional personal-data patterns. Opaque files may be allowed by default or explicitly blocked. AHV/IBAN and confidential-file detection accuracy were not tested. Sources: librechat:librechat:r4-content-filter, librechat:librechat:r4-pii-runtime, librechat:librechat:r4-patterns, librechat:librechat:r4-private-submission, librechat:librechat:r4-filter-middleware. - pii_action — Is it redacted, masked, blocked or only warned? [partial]: Configured matches can block or audit. Redaction transforms only fresh, text-only interactive Agent submissions carrying clientRequestId and a valid CREDS_KEY; edits, resumes, files, quotes and other endpoints retain blocking checks. Original text is encrypted and unindexed, rather than discarded. No running deployment was tested. Sources: librechat:librechat:r4-content-filter, librechat:librechat:r4-pii-runtime, librechat:librechat:r4-patterns, librechat:librechat:r4-private-submission, librechat:librechat:r4-filter-middleware. - custom_patterns — Can admins add custom patterns or terms? [supported]: Operators can add labelled custom RE2JS regular expressions in base librechat.yaml, with optional categories. Database/role/group/user overrides cannot modify the policy; all replicas must load matching configuration. Invalid rules are rejected on configuration load. Sources: librechat:librechat:r4-content-filter, librechat:librechat:r4-patterns, librechat:librechat:r4-filter-middleware. - dlp_integration — Does it integrate with an existing DLP product? [unknown]: No first-party Purview/Netskope or other external DLP connector was established from the content-filter guide, current public repository tree and targeted official-site search. Configurable local patterns are not evidence of an external enterprise DLP integration. - filter_location — Where does filtering run? [supported]: Inspected content-filter middleware and rule evaluation run in the customer-operated LibreChat API backend. Optional extraction/OCR and inference providers remain separate data flows; no named deployment region or traffic path was tested. Sources: librechat:librechat:r4-pii-runtime, librechat:librechat:r4-filter-middleware. Checked 2026-10-26; agent cross-checked 2026-10-26; check again by 2026-12-25. Sources: librechat:librechat:r4-content-filter, librechat:librechat:r4-pii-runtime, librechat:librechat:r4-patterns, librechat:librechat:r4-private-submission, librechat:librechat:r4-filter-middleware, librechat:librechat:r4-config-1318 ### ADM-08: AI policy and guardrails Judgment rule: Record organisation-wide system instructions, acceptable-use notices and acknowledgement, content filters, topic restrictions, AI-output labelling, and admin visibility of policy violations. Separate enforced policy from instructions the model is expected to follow. #### Dust Verification status: Agent cross-checked (agent-cross-checked) Support: partial Agents have persistent instructions and workspace governance; broad enforced content-policy administration is unverified. Availability: generally-available Delivery: native Scope: surfaces=Dust web workspace, Dust EU web workspace; version=Public source pass 2026-10-21; no logged-in deployment observed; plan=Business credit-priced plan; Enterprise or legacy differences are named per answer; add_ons=none stated; deployment_mode=multi-tenant-saas; region=US/global or EU workspace as named; EU-hosted-model restriction must be enabled separately; os=not specified. Limitations: - Documented findings only; unestablished atomic questions remain explicit gaps. - org_instructions — Can admins set organisation-wide instructions? [partial]: Admins/builders configure agent instructions; an enforced single organization-wide instruction applied to every agent/message was not established. Sources: dust:dust:builder, dust:dust:governance. - usage_notice — Can users be required to accept a usage policy? [unknown]: Can users be required to accept a usage policy? Agent, governance and security pages were checked. Mandatory usage-policy acknowledgement and administrator policy-violation reporting were not established. - content_filters — Are content filters configurable? [partial]: Image Generation documents NSFW/celebrity safety filtering. Custom organization-wide topic/content rules were not established. Sources: dust:dust:images. - violation_reporting — Are policy hits visible to admins? [unknown]: Are policy hits visible to admins? Agent, governance and security pages were checked. Mandatory usage-policy acknowledgement and administrator policy-violation reporting were not established. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: dust:dust:builder, dust:dust:governance, dust:dust:images #### LibreChat Verification status: Agent cross-checked (agent-cross-checked) Support: partial Operators can configure prompts and usage terms; abuse moderation is distinct from semantic content-policy enforcement. Availability: generally-available Delivery: native Scope: surfaces=Customer-operated web application and server; version=Public documentation retrieved 2026-10-21; inspected source v0.8.8 / e8f3be08623663d4ad7f7241e693c94469b63bb0; plan=MIT customer-operated LibreChat; optional separately deployed services identified per answer; add_ons=none stated; deployment_mode=on-premise; region=Customer-selected infrastructure; model, OCR, search, observability and support services separately scoped; os=not specified. Limitations: - org_instructions: Enforced model-spec presets and Agent instructions can guide behavior; model-followed instructions are not guaranteed policy enforcement. - content_filters: Automated moderation scores rate/authentication/message abuse and can ban users. Semantic topic or PII policy coverage was not established. - violation_reporting: Abuse scoring is documented, but a complete organization content-policy violation dashboard and reporting fixture were not established. - org_instructions — Can admins set organisation-wide instructions? [partial]: Enforced model-spec presets and Agent instructions can guide behavior; model-followed instructions are not guaranteed policy enforcement. Sources: librechat:librechat:modelspec, librechat:librechat:agents. - usage_notice — Can users be required to accept a usage policy? [supported]: Interface configuration includes custom terms of service and acceptance settings. Sources: librechat:librechat:interface. - content_filters — Are content filters configurable? [partial]: Automated moderation scores rate/authentication/message abuse and can ban users. Semantic topic or PII policy coverage was not established. Sources: librechat:librechat:mod. - violation_reporting — Are policy hits visible to admins? [partial]: Abuse scoring is documented, but a complete organization content-policy violation dashboard and reporting fixture were not established. Sources: librechat:librechat:mod. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: librechat:librechat:modelspec, librechat:librechat:agents, librechat:librechat:interface, librechat:librechat:mod ## Data protection and sovereignty ### DATA-01: Hosting location and data residency Judgment rule: Record country or region separately for chat storage, files and indexes, model inference, backups, logs and telemetry, and support access. Name the cloud provider. Record whether residency is default, a paid option or plan-dependent. "EU data boundary" programmes are recorded with their documented exceptions. #### Dust Verification status: Agent cross-checked (agent-cross-checked) Support: partial EU workspace and EU model controls are distinct from general hosting and global subprocessing. Availability: generally-available Delivery: native Scope: surfaces=Dust web workspace, Dust EU web workspace; version=Public source pass 2026-10-21; no logged-in deployment observed; plan=Business credit-priced plan; Enterprise or legacy differences are named per answer; add_ons=none stated; deployment_mode=multi-tenant-saas; region=US/global or EU workspace as named; EU-hosted-model restriction must be enabled separately; os=not specified. Limitations: - Documented findings only; unestablished atomic questions remain explicit gaps. - storage_region — Where are chats, files and indexes stored? [partial]: Pricing/security advertise EU or US workspace hosting. Exact cloud provider, storage country and index location were not established. Sources: dust:dust:pricing, dust:dust:security. - inference_region — Where do model calls run, per provider? [supported]: EU credit-priced workspaces using Dust-hosted models can enable EU-hosted models only after all active agents comply. BYOK is outside this setting; not every model is eligible. Sources: dust:dust:eu-models. - backups_logs_region — Where are backups, logs and telemetry? [unknown]: Where are backups, logs and telemetry? Hosting, EU-model and privacy pages were checked. Backup/log/telemetry countries, customer-content support-access geography and Swiss hosting were not established. - support_access_region — From where can vendor staff access data? [partial]: The controller privacy policy permits non-EEA access/processing including the US; it does not establish customer-content support access by staff country. Sources: dust:dust:privacy. - swiss_hosting — Is Swiss hosting available? [unknown]: Is Swiss hosting available? Hosting, EU-model and privacy pages were checked. Backup/log/telemetry countries, customer-content support-access geography and Swiss hosting were not established. - residency_plan — Which plan or option provides residency? [supported]: Business and Enterprise list EU/US residency; EU-only model restrictions additionally require EU region and credit pricing with Dust model hosting. Sources: dust:dust:pricing, dust:dust:eu-models. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: dust:dust:pricing, dust:dust:security, dust:dust:eu-models, dust:dust:privacy #### LibreChat Verification status: Agent cross-checked (agent-cross-checked) Support: partial Customer operators choose storage and service locations; external model, OCR, search and code services need separate residency decisions. Availability: generally-available Delivery: native Scope: surfaces=Customer-operated web application and server; version=Public documentation retrieved 2026-10-21; inspected source v0.8.8 / e8f3be08623663d4ad7f7241e693c94469b63bb0; plan=MIT customer-operated LibreChat; optional separately deployed services identified per answer; add_ons=none stated; deployment_mode=on-premise; region=Customer-selected infrastructure; model, OCR, search, observability and support services separately scoped; os=not specified. Limitations: - backups_logs_region: Logs and mounted data reside in the configured deployment; backup, telemetry and optional tracing geographies depend on operator choices and were not audited. - support_access_region: From where can vendor staff access data? No vendor or integrator support-access country restriction was reviewed for this customer-operated deployment. - swiss_hosting: A customer may run the software on Swiss infrastructure; no bundled Swiss managed inference/support commitment was established. - storage_region — Where are chats, files and indexes stored? [supported]: The reference self-hosted stack keeps application data in configured MongoDB/files and retrieval data in PGVector. Its country is chosen by the deployer. Sources: librechat:librechat:compose-pin, librechat:librechat:remote. - inference_region — Where do model calls run, per provider? [supported]: Customer-selected cloud endpoints/regions or local endpoints determine inference; hosting LibreChat in Switzerland does not place every external model there. Sources: librechat:librechat:custom, librechat:librechat:bedrock, librechat:librechat:vertex. - backups_logs_region — Where are backups, logs and telemetry? [partial]: Logs and mounted data reside in the configured deployment; backup, telemetry and optional tracing geographies depend on operator choices and were not audited. Sources: librechat:librechat:logging, librechat:librechat:compose-pin. - support_access_region — From where can vendor staff access data? [unknown]: support_access_region: From where can vendor staff access data? No vendor or integrator support-access country restriction was reviewed for this customer-operated deployment. - swiss_hosting — Is Swiss hosting available? [partial]: A customer may run the software on Swiss infrastructure; no bundled Swiss managed inference/support commitment was established. Sources: librechat:librechat:remote, librechat:librechat:license. - residency_plan — Which plan or option provides residency? [supported]: Customer-controlled placement is a self-hosted deployment property, not a separately priced residency licence. Sources: librechat:librechat:remote, librechat:librechat:license. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: librechat:librechat:compose-pin, librechat:librechat:remote, librechat:librechat:custom, librechat:librechat:bedrock, librechat:librechat:vertex, librechat:librechat:logging, librechat:librechat:license ### DATA-02: Deployment options Judgment rule: Record multi-tenant SaaS, dedicated single-tenant instance, deployment in the customer's cloud account, on-premise and air-gapped as separate options with plan and availability. Record which features are missing in each mode, who operates it, and the documented seat or contract threshold for each mode, which can differ from the enterprise plan threshold (e.g. dedicated SaaS from 2,000 seats, own cloud from 5,000). A local-network connector or gateway that reaches on-prem data from a vendor-hosted platform is KNOW-03, not on-premise deployment. #### Dust Verification status: Agent cross-checked (agent-cross-checked) Support: partial Hosted Business and Enterprise single-tenant offers are public; supported customer-operated whole-platform modes are unverified. Availability: generally-available Delivery: native Scope: surfaces=Dust web workspace, Dust EU web workspace, Dust Enterprise single-tenant deployment; version=Public source pass 2026-10-21; no logged-in deployment observed; plan=Business credit-priced plan; Enterprise or legacy differences are named per answer; add_ons=none stated; deployment_mode=not specified; region=US/global or EU workspace as named; EU-hosted-model restriction must be enabled separately; os=not specified. Limitations: - Documented findings only; unestablished atomic questions remain explicit gaps. - multi_tenant_saas — Is a shared SaaS offered? [partial]: Business is a self-serve hosted workspace; exact shared-infrastructure isolation architecture is not documented in the checked pages. Sources: dust:dust:pricing, dust:dust:privacy. - dedicated_instance — Is a dedicated single-tenant instance offered, from what size? [supported]: Enterprise pricing advertises single-tenant deployment. Numerical seat/spend minimum, supported countries and operating terms were not established. Sources: dust:dust:pricing. - customer_cloud — Can it run in the customer's own cloud account, from what size? [unknown]: Can it run in the customer's own cloud account, from what size? Pricing, security and the public repository were checked. Customer-cloud, whole-platform on-premise/air-gapped support and their thresholds were not established; MIT source alone does not promise a supported product deployment. - on_premise — Can the whole platform run on customer infrastructure, from what size? [unknown]: Can the whole platform run on customer infrastructure, from what size? Pricing, security and the public repository were checked. Customer-cloud, whole-platform on-premise/air-gapped support and their thresholds were not established; MIT source alone does not promise a supported product deployment. - air_gapped — Can it run without internet access? [unknown]: Can it run without internet access? Pricing, security and the public repository were checked. Customer-cloud, whole-platform on-premise/air-gapped support and their thresholds were not established; MIT source alone does not promise a supported product deployment. - mode_feature_gaps — Which features are missing outside SaaS? [partial]: Business/Enterprise packaging differs in connectors, MCP, Spaces, APIs and governance. No customer-operated hosting feature-parity contract was established. Sources: dust:dust:pricing. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: dust:dust:pricing, dust:dust:privacy #### LibreChat Verification status: Agent cross-checked (agent-cross-checked) Support: partial The project documents local and customer-cloud installs. A free public demo is separate from a production managed service. Availability: generally-available Delivery: native Scope: surfaces=Customer-operated web application and server, LibreChat public evaluation demo; version=Public documentation retrieved 2026-10-21; inspected source v0.8.8 / e8f3be08623663d4ad7f7241e693c94469b63bb0; plan=MIT customer-operated LibreChat; optional separately deployed services identified per answer; add_ons=none stated; deployment_mode=on-premise; region=Customer-selected infrastructure; model, OCR, search, observability and support services separately scoped; os=not specified. Limitations: - multi_tenant_saas: The project operates a free public evaluation Demo, not a reviewed production business SaaS offer. Its terms permit data cleanups and disclaim persistence. - dedicated_instance: Is a dedicated single-tenant instance offered, from what size? A vendor-operated dedicated production service and its contractual size floor were not established. - air_gapped: Local chat/embedding routes exist, but a complete disconnected installation and its full dependency behavior were not verified. - mode_feature_gaps: External APIs, OCR/search services and separate code runtimes need their own availability/networking; there is no verified SaaS/on-prem feature-parity matrix. - multi_tenant_saas — Is a shared SaaS offered? [partial]: The project operates a free public evaluation Demo, not a reviewed production business SaaS offer. Its terms permit data cleanups and disclaim persistence. Sources: librechat:librechat:demo, librechat:librechat:firstchat. - dedicated_instance — Is a dedicated single-tenant instance offered, from what size? [unknown]: dedicated_instance: Is a dedicated single-tenant instance offered, from what size? A vendor-operated dedicated production service and its contractual size floor were not established. - customer_cloud — Can it run in the customer's own cloud account, from what size? [supported]: Remote hosting documentation describes customer-selected servers/cloud hosting without a software seat minimum. Sources: librechat:librechat:remote, librechat:librechat:license. - on_premise — Can the whole platform run on customer infrastructure, from what size? [supported]: Docker/npm installation can run the application and dependencies on customer infrastructure. Sources: librechat:librechat:local, librechat:librechat:compose-pin. - air_gapped — Can it run without internet access? [partial]: Local chat/embedding routes exist, but a complete disconnected installation and its full dependency behavior were not verified. Sources: librechat:librechat:custom, librechat:librechat:ragcfg. - mode_feature_gaps — Which features are missing outside SaaS? [partial]: External APIs, OCR/search services and separate code runtimes need their own availability/networking; there is no verified SaaS/on-prem feature-parity matrix. Sources: librechat:librechat:ocr, librechat:librechat:web, librechat:librechat:code. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: librechat:librechat:demo, librechat:librechat:firstchat, librechat:librechat:remote, librechat:librechat:license, librechat:librechat:local, librechat:librechat:compose-pin, librechat:librechat:custom, librechat:librechat:ragcfg, librechat:librechat:ocr, librechat:librechat:web, librechat:librechat:code ### DATA-03: Training use and provider retention Judgment rule: Record whether customer content trains vendor or provider models, by plan and default; zero-data-retention arrangements with model providers; abuse-monitoring retention and human review; and the model provider's terms that apply. A "no training" claim on a marketing page is `vendor-claimed` until the contract or DPA says so. #### Dust Verification status: Agent cross-checked (agent-cross-checked) Support: partial Dust publishes no-training and provider-retention claims; the customer DPA was not retrieved. Availability: generally-available Delivery: native Scope: surfaces=Dust web workspace, Dust EU web workspace; version=Public source pass 2026-10-21; no logged-in deployment observed; plan=Business credit-priced plan; Enterprise or legacy differences are named per answer; add_ons=none stated; deployment_mode=multi-tenant-saas; region=US/global or EU workspace as named; EU-hosted-model restriction must be enabled separately; os=not specified. Limitations: - Controller privacy-policy scope is distinct from processor DPA coverage. - vendor_training — Does the vendor train on customer content? [partial]: The security page claims customer data is never used for model training. This remains vendor-claimed because no customer DPA/contract clause was retrieved. Sources: dust:dust:security. - provider_training — Do model providers train on it under this product's contracts? [partial]: The privacy policy names OpenAI, Anthropic, Mistral, Google and Fireworks as prohibited from training on customer/personal content; BYOK arrangements need separate confirmation. Sources: dust:dust:training-policy. - provider_retention — How long do model providers keep prompts, including for abuse monitoring? [partial]: The policy states providers do not log Customer Content for human review or save it to disk. Exception/abuse-retention periods were not established from customer contracts. Sources: dust:dust:training-policy. - zero_retention — Is zero data retention available, and on which plan? [partial]: Security and privacy pages assert provider zero retention, but applicable plan/model exceptions were not contract-verified. Sources: dust:dust:security, dust:dust:training-policy. - human_review — Can provider staff review content? [partial]: The published policy says listed providers do not log Customer Content for human review. This is a vendor assertion, not an independently reviewed model-provider contract. Sources: dust:dust:training-policy. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: dust:dust:security, dust:dust:training-policy #### LibreChat Verification status: Agent cross-checked (agent-cross-checked) Support: partial The customer configures model contracts and data paths; a universal no-training or provider-retention promise was not established. Availability: generally-available Delivery: native Scope: surfaces=Customer-operated web application and server; version=Public documentation retrieved 2026-10-21; inspected source v0.8.8 / e8f3be08623663d4ad7f7241e693c94469b63bb0; plan=MIT customer-operated LibreChat; optional separately deployed services identified per answer; add_ons=none stated; deployment_mode=on-premise; region=Customer-selected infrastructure; model, OCR, search, observability and support services separately scoped; os=not specified. Limitations: - vendor_training: The self-hosted application stores data in the customer’s configured stack; this architecture is not a contractual promise covering every optional external service. - provider_training: Do model providers train on it under this product's contracts? No specific customer provider agreement, abuse retention/human-review period or zero-retention entitlement was reviewed. Website and Demo privacy policies do not govern arbitrary self-hosted providers. - provider_retention: How long do model providers keep prompts, including for abuse monitoring? No specific customer provider agreement, abuse retention/human-review period or zero-retention entitlement was reviewed. Website and Demo privacy policies do not govern arbitrary self-hosted providers. - zero_retention: Is zero data retention available, and on which plan? No specific customer provider agreement, abuse retention/human-review period or zero-retention entitlement was reviewed. Website and Demo privacy policies do not govern arbitrary self-hosted providers. - human_review: Can provider staff review content? No specific customer provider agreement, abuse retention/human-review period or zero-retention entitlement was reviewed. Website and Demo privacy policies do not govern arbitrary self-hosted providers. - vendor_training — Does the vendor train on customer content? [partial]: The self-hosted application stores data in the customer’s configured stack; this architecture is not a contractual promise covering every optional external service. Sources: librechat:librechat:compose-pin, librechat:librechat:firstchat. - provider_training — Do model providers train on it under this product's contracts? [unknown]: provider_training: Do model providers train on it under this product's contracts? No specific customer provider agreement, abuse retention/human-review period or zero-retention entitlement was reviewed. Website and Demo privacy policies do not govern arbitrary self-hosted providers. - provider_retention — How long do model providers keep prompts, including for abuse monitoring? [unknown]: provider_retention: How long do model providers keep prompts, including for abuse monitoring? No specific customer provider agreement, abuse retention/human-review period or zero-retention entitlement was reviewed. Website and Demo privacy policies do not govern arbitrary self-hosted providers. - zero_retention — Is zero data retention available, and on which plan? [unknown]: zero_retention: Is zero data retention available, and on which plan? No specific customer provider agreement, abuse retention/human-review period or zero-retention entitlement was reviewed. Website and Demo privacy policies do not govern arbitrary self-hosted providers. - human_review — Can provider staff review content? [unknown]: human_review: Can provider staff review content? No specific customer provider agreement, abuse retention/human-review period or zero-retention entitlement was reviewed. Website and Demo privacy policies do not govern arbitrary self-hosted providers. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: librechat:librechat:compose-pin, librechat:librechat:firstchat ### DATA-04: Retention and deletion Judgment rule: Record default retention for chats, files, indexes and logs; admin-configured retention; user deletion; deletion after contract end; and deletion confirmation. #### Dust Verification status: Agent cross-checked (agent-cross-checked) Support: partial Some user deletion and memory retention behavior is documented; tenant-wide retention schedules remain open. Availability: generally-available Delivery: native Scope: surfaces=Dust web workspace, Dust EU web workspace; version=Public source pass 2026-10-21; no logged-in deployment observed; plan=Business credit-priced plan; Enterprise or legacy differences are named per answer; add_ons=none stated; deployment_mode=multi-tenant-saas; region=US/global or EU workspace as named; EU-hosted-model restriction must be enabled separately; os=not specified. Limitations: - Documented findings only; unestablished atomic questions remain explicit gaps. - default_retention — What is retained and for how long by default? [partial]: Agent memories persist until manually deleted. This does not define default retention for all conversations, files, indexes or logs. Sources: dust:dust:memory. - admin_retention — Can admins set retention periods? [partial]: Enterprise pricing offers custom data retention; periods, deletion mechanics and exceptions were not retrieved. Sources: dust:dust:pricing. - user_deletion — Can users delete their chats and files permanently? [partial]: Users can delete conversations and agent memories. Backup erasure timing and all attached-file deletion behavior were not established. Sources: dust:dust:delete-chat, dust:dust:memory. - contract_end_deletion — What is deleted at contract end, and when? [unknown]: What is deleted at contract end, and when? Conversation/memory docs, Enterprise pricing and unreadable legal entry points were checked. Contract-end erasure timing and complete storage/index/log retention defaults remain unknown. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: dust:dust:memory, dust:dust:pricing, dust:dust:delete-chat #### LibreChat Verification status: Agent cross-checked (agent-cross-checked) Support: partial Normal chats have no default expiry; temporary chats default to 30 days and admins can configure broader expiry modes. Availability: generally-available Delivery: native Scope: surfaces=Customer-operated web application and server; version=Public documentation retrieved 2026-10-21; inspected source v0.8.8 / e8f3be08623663d4ad7f7241e693c94469b63bb0; plan=MIT customer-operated LibreChat; optional separately deployed services identified per answer; add_ons=none stated; deployment_mode=on-premise; region=Customer-selected infrastructure; model, OCR, search, observability and support services separately scoped; os=not specified. Limitations: - user_deletion: Users can delete conversations/data and retention sweeps remove files. Copies, backups and external provider deletion were not end-to-end verified. - contract_end_deletion: What is deleted at contract end, and when? No commercial contract-end purge/confirmation deadline was reviewed; the Demo’s retention terms are a separate offer. - default_retention — What is retained and for how long by default? [supported]: The default temporary retention is 720 hours; normal chats persist unless configured under all/ephemeral modes. Memories do not expire with chats. Sources: librechat:librechat:retention. - admin_retention — Can admins set retention periods? [supported]: Admins set all or ephemeral retention, regular/temporary deadlines and Agent file retention. New settings apply on later saves rather than rewriting every existing record immediately. Sources: librechat:librechat:retention. - user_deletion — Can users delete their chats and files permanently? [partial]: Users can delete conversations/data and retention sweeps remove files. Copies, backups and external provider deletion were not end-to-end verified. Sources: librechat:librechat:retention, librechat:librechat:settings. - contract_end_deletion — What is deleted at contract end, and when? [unknown]: contract_end_deletion: What is deleted at contract end, and when? No commercial contract-end purge/confirmation deadline was reviewed; the Demo’s retention terms are a separate offer. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: librechat:librechat:retention, librechat:librechat:settings ### DATA-05: Sub-processors and foreign-law exposure Judgment rule: Record the published sub-processor list with countries, change-notification rules, the vendor's and parent company's jurisdiction, transfer mechanisms (EU-US Data Privacy Framework, Swiss-US DPF, SCCs) and the vendor's documented position on foreign government access requests (e.g. US CLOUD Act). Record facts, not a legal assessment. #### Dust Verification status: Agent cross-checked (agent-cross-checked) Support: partial Dust identifies a French operating entity; customer subprocessor inventory and notice rules remain unverified. Availability: generally-available Delivery: native Scope: surfaces=Dust web workspace, Dust EU web workspace; version=Public source pass 2026-10-21; no logged-in deployment observed; plan=Business credit-priced plan; Enterprise or legacy differences are named per answer; add_ons=none stated; deployment_mode=multi-tenant-saas; region=US/global or EU workspace as named; EU-hosted-model restriction must be enabled separately; os=not specified. Limitations: - Documented findings only; unestablished atomic questions remain explicit gaps. - subprocessor_list — Is a dated sub-processor list published? [unknown]: Is a dated sub-processor list published? The privacy policy refers to a Sub-Processors list, but the official Terms/Notion and trust-center routes were not readable. A dated customer-processor list, complete countries and change-notification rules were not established. - subprocessor_countries — Which countries do sub-processors operate in? [unknown]: Which countries do sub-processors operate in? The privacy policy refers to a Sub-Processors list, but the official Terms/Notion and trust-center routes were not readable. A dated customer-processor list, complete countries and change-notification rules were not established. - change_notification — Are customers notified of sub-processor changes? [unknown]: Are customers notified of sub-processor changes? The privacy policy refers to a Sub-Processors list, but the official Terms/Notion and trust-center routes were not readable. A dated customer-processor list, complete countries and change-notification rules were not established. - parent_jurisdiction — Which jurisdiction governs the vendor and its parent? [partial]: The public policy identifies Permutation Labs at a Paris, France address. A separate parent company or controlling entity was not established. Sources: dust:dust:privacy. - transfer_mechanism — Which transfer mechanism applies to non-EU/CH processing? [partial]: For its controller processing, the privacy policy describes adequacy or EU SCC safeguards for non-EU transfers, including the US. Customer processor transfer terms were not retrieved. Sources: dust:dust:privacy. - government_access_policy — Is there a published government-access or transparency policy? [partial]: The controller policy allows legally required disclosure to authorities; a dedicated government-request transparency policy for customer content was not established. Sources: dust:dust:privacy. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: dust:dust:privacy #### LibreChat Verification status: Agent cross-checked (agent-cross-checked) Support: partial ClickHouse’s acquisition is sourced; the self-hosted deployment still selects its own external processors and contracts. Availability: generally-available Delivery: native Scope: surfaces=Customer-operated web application and server; version=Public documentation retrieved 2026-10-21; inspected source v0.8.8 / e8f3be08623663d4ad7f7241e693c94469b63bb0; plan=MIT customer-operated LibreChat; optional separately deployed services identified per answer; add_ons=none stated; deployment_mode=on-premise; region=Customer-selected infrastructure; model, OCR, search, observability and support services separately scoped; os=not specified. Limitations: - subprocessor_list: Is a dated sub-processor list published? No project-wide dated sub-processor/country list, change-notification schedule, transfer instrument or government-access policy was established for arbitrary customer-operated deployments. Website/Demo policies are separate scopes. - subprocessor_countries: Which countries do sub-processors operate in? No project-wide dated sub-processor/country list, change-notification schedule, transfer instrument or government-access policy was established for arbitrary customer-operated deployments. Website/Demo policies are separate scopes. - change_notification: Are customers notified of sub-processor changes? No project-wide dated sub-processor/country list, change-notification schedule, transfer instrument or government-access policy was established for arbitrary customer-operated deployments. Website/Demo policies are separate scopes. - parent_jurisdiction: ClickHouse, Inc. announced the LibreChat acquisition on 2025-11-04 from San Francisco and brought founder/team into ClickHouse. A self-hosted customer contracting entity/governing jurisdiction was not established from that announcement. - transfer_mechanism: Which transfer mechanism applies to non-EU/CH processing? No project-wide dated sub-processor/country list, change-notification schedule, transfer instrument or government-access policy was established for arbitrary customer-operated deployments. Website/Demo policies are separate scopes. - government_access_policy: Is there a published government-access or transparency policy? No project-wide dated sub-processor/country list, change-notification schedule, transfer instrument or government-access policy was established for arbitrary customer-operated deployments. Website/Demo policies are separate scopes. - subprocessor_list — Is a dated sub-processor list published? [unknown]: subprocessor_list: Is a dated sub-processor list published? No project-wide dated sub-processor/country list, change-notification schedule, transfer instrument or government-access policy was established for arbitrary customer-operated deployments. Website/Demo policies are separate scopes. - subprocessor_countries — Which countries do sub-processors operate in? [unknown]: subprocessor_countries: Which countries do sub-processors operate in? No project-wide dated sub-processor/country list, change-notification schedule, transfer instrument or government-access policy was established for arbitrary customer-operated deployments. Website/Demo policies are separate scopes. - change_notification — Are customers notified of sub-processor changes? [unknown]: change_notification: Are customers notified of sub-processor changes? No project-wide dated sub-processor/country list, change-notification schedule, transfer instrument or government-access policy was established for arbitrary customer-operated deployments. Website/Demo policies are separate scopes. - parent_jurisdiction — Which jurisdiction governs the vendor and its parent? [partial]: ClickHouse, Inc. announced the LibreChat acquisition on 2025-11-04 from San Francisco and brought founder/team into ClickHouse. A self-hosted customer contracting entity/governing jurisdiction was not established from that announcement. Sources: librechat:librechat:owner. - transfer_mechanism — Which transfer mechanism applies to non-EU/CH processing? [unknown]: transfer_mechanism: Which transfer mechanism applies to non-EU/CH processing? No project-wide dated sub-processor/country list, change-notification schedule, transfer instrument or government-access policy was established for arbitrary customer-operated deployments. Website/Demo policies are separate scopes. - government_access_policy — Is there a published government-access or transparency policy? [unknown]: government_access_policy: Is there a published government-access or transparency policy? No project-wide dated sub-processor/country list, change-notification schedule, transfer instrument or government-access policy was established for arbitrary customer-operated deployments. Website/Demo policies are separate scopes. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: librechat:librechat:owner ### DATA-06: Contracts and regulatory coverage Judgment rule: Record DPA availability and form (click-through, negotiable), reference to Swiss revDSG and GDPR, governing law and court, EU AI Act documentation (provider or deployer role, transparency materials), and sector-specific materials (FINMA outsourcing, banking secrecy, public sector). No compliance verdict. #### Dust Verification status: Agent cross-checked (agent-cross-checked) Support: partial Enterprise offers custom MSA/DPA terms, while the underlying legal materials were inaccessible. Availability: generally-available Delivery: native Scope: surfaces=Dust web workspace, Dust EU web workspace; version=Public source pass 2026-10-21; no logged-in deployment observed; plan=Business credit-priced plan; Enterprise or legacy differences are named per answer; add_ons=none stated; deployment_mode=multi-tenant-saas; region=US/global or EU workspace as named; EU-hosted-model restriction must be enabled separately; os=not specified. Limitations: - Documented findings only; unestablished atomic questions remain explicit gaps. - dpa — Is a DPA available, and is it negotiable? [partial]: Enterprise pricing explicitly offers custom MSA/DPA terms. The form, sign-up incorporation and negotiability boundaries were not inspected. Sources: dust:dust:pricing. - revdsg_gdpr — Does the DPA cover revDSG and GDPR? [partial]: Marketing and privacy materials reference GDPR; explicit Swiss revDSG coverage in the customer DPA was not established. Sources: dust:dust:security, dust:dust:privacy. - governing_law — Which law and court govern the contract? [unknown]: Which law and court govern the contract? Official Terms & Policies redirects to a Notion page that returned no readable terms; trust.dust.com also exposed no report text. Governing law/court and EU AI Act technical/role materials were not established. - ai_act_materials — Is EU AI Act documentation provided? [unknown]: Is EU AI Act documentation provided? Official Terms & Policies redirects to a Notion page that returned no readable terms; trust.dust.com also exposed no report text. Governing law/court and EU AI Act technical/role materials were not established. - sector_materials — Are FINMA or other sector materials provided? [partial]: Security marketing says the service enables HIPAA compliance. A BAA, FINMA or banking-secrecy outsourcing pack was not retrieved. Sources: dust:dust:security. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: dust:dust:pricing, dust:dust:security, dust:dust:privacy #### LibreChat Verification status: No public information found (not-published) Support: not-published We found no public information on this. A missing finding is not evidence that the product lacks the capability. ### DATA-07: Certifications and attestations Judgment rule: Record each certificate or report separately: ISO 27001, 27017, 27018, 27701, 42001; SOC 2 Type I/II with period; BSI C5; TISAX; national schemes. Record issuer, scope (which product and entity), validity and access route. A cloud provider's certificate does not certify the product. #### Dust Verification status: Agent cross-checked (agent-cross-checked) Support: partial SOC 2 Type II is claimed publicly, without a verified report or certificate. Availability: generally-available Delivery: native Scope: surfaces=Dust web workspace, Dust EU web workspace; version=Public source pass 2026-10-21; no logged-in deployment observed; plan=Business credit-priced plan; Enterprise or legacy differences are named per answer; add_ons=none stated; deployment_mode=multi-tenant-saas; region=US/global or EU workspace as named; EU-hosted-model restriction must be enabled separately; os=not specified. Limitations: - Documented findings only; unestablished atomic questions remain explicit gaps. - iso_27001 — ISO 27001 with scope and validity? [unknown]: ISO 27001 with scope and validity? Security/pricing and the trust center were checked. ISO 27001, C5, other scoped certificates and independent penetration-test summaries were not established; no evidence is certificate-verified. - soc2 — SOC 2 type, period and covered service? [partial]: Dust security/pricing claim SOC 2 Type II. Auditor, reporting period and exact service scope were not verified. Sources: dust:dust:security, dust:dust:pricing. - c5 — BSI C5 attestation? [unknown]: BSI C5 attestation? Security/pricing and the trust center were checked. ISO 27001, C5, other scoped certificates and independent penetration-test summaries were not established; no evidence is certificate-verified. - other_certifications — Which other certificates (27701, 42001, TISAX)? [unknown]: Which other certificates (27701, 42001, TISAX)? Security/pricing and the trust center were checked. ISO 27001, C5, other scoped certificates and independent penetration-test summaries were not established; no evidence is certificate-verified. - report_access — How can customers obtain reports? [partial]: The security page directs customers to trust.dust.com; underlying reports were not accessible in this public source pass. Sources: dust:dust:security. - pentest — Are independent penetration test summaries available? [unknown]: Are independent penetration test summaries available? Security/pricing and the trust center were checked. ISO 27001, C5, other scoped certificates and independent penetration-test summaries were not established; no evidence is certificate-verified. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: dust:dust:security, dust:dust:pricing #### LibreChat Verification status: No public information found (not-published) Support: not-published We found no public information on this. A missing finding is not evidence that the product lacks the capability. ### DATA-08: Encryption and key control Judgment rule: Record encryption at rest and in transit, customer-managed keys (BYOK/HYOK), key location, and private networking (private link, IP allow-lists, VPN). #### Dust Verification status: Agent cross-checked (agent-cross-checked) Support: partial Dust claims AES-256/TLS and advertises Enterprise single tenancy; customer key custody is unverified. Availability: generally-available Delivery: native Scope: surfaces=Dust web workspace, Dust EU web workspace; version=Public source pass 2026-10-21; no logged-in deployment observed; plan=Business credit-priced plan; Enterprise or legacy differences are named per answer; add_ons=none stated; deployment_mode=multi-tenant-saas; region=US/global or EU workspace as named; EU-hosted-model restriction must be enabled separately; os=not specified. Limitations: - Documented findings only; unestablished atomic questions remain explicit gaps. - encryption_at_rest — Is data encrypted at rest? [supported]: The security page states AES-256 at rest and TLS in transit; this is a vendor claim. Sources: dust:dust:security. - customer_managed_keys — Can the customer manage or hold the keys? [unknown]: Can the customer manage or hold the keys? Security, pricing and network/MCP docs were checked. Customer-managed/held data-encryption keys and key-location guarantees were not established. Provider API BYOK is not encryption-key custody. - private_networking — Is private connectivity or IP restriction available? [partial]: Custom MCP servers can restrict access and use documented static egress on verified domains. This is tool connectivity, not proof of private connectivity to the whole service. Sources: dust:dust:tools. - tenant_isolation — How are customers separated in shared SaaS? [partial]: Enterprise offers single tenancy; shared Business tenancy and lower-level isolation mechanics were not established. Sources: dust:dust:pricing. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: dust:dust:security, dust:dust:tools, dust:dust:pricing #### LibreChat Verification status: Agent cross-checked (agent-cross-checked) Support: partial Network and storage security depend on the customer’s deployment; application authentication does not certify encryption of all stored content. Availability: generally-available Delivery: native Scope: surfaces=Customer-operated web application and server; version=Public documentation retrieved 2026-10-21; inspected source v0.8.8 / e8f3be08623663d4ad7f7241e693c94469b63bb0; plan=MIT customer-operated LibreChat; optional separately deployed services identified per answer; add_ons=none stated; deployment_mode=on-premise; region=Customer-selected infrastructure; model, OCR, search, observability and support services separately scoped; os=not specified. Limitations: - encryption_at_rest: Is data encrypted at rest? No default at-rest encryption guarantee covering MongoDB/files/PGVector or audited multi-tenant managed-service isolation was established from the installation guides. - customer_managed_keys: Operators own infrastructure and application secrets; a documented product HYOK/customer-managed content-encryption service was not verified. - tenant_isolation: How are customers separated in shared SaaS? No default at-rest encryption guarantee covering MongoDB/files/PGVector or audited multi-tenant managed-service isolation was established from the installation guides. - encryption_at_rest — Is data encrypted at rest? [unknown]: encryption_at_rest: Is data encrypted at rest? No default at-rest encryption guarantee covering MongoDB/files/PGVector or audited multi-tenant managed-service isolation was established from the installation guides. - customer_managed_keys — Can the customer manage or hold the keys? [partial]: Operators own infrastructure and application secrets; a documented product HYOK/customer-managed content-encryption service was not verified. Sources: librechat:librechat:auth, librechat:librechat:compose-pin. - private_networking — Is private connectivity or IP restriction available? [supported]: Operators deploy the stack on their own network and configure reverse proxies/HTTPS for remote access. Sources: librechat:librechat:remote, librechat:librechat:local. - tenant_isolation — How are customers separated in shared SaaS? [unknown]: tenant_isolation: How are customers separated in shared SaaS? No default at-rest encryption guarantee covering MongoDB/files/PGVector or audited multi-tenant managed-service isolation was established from the installation guides. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: librechat:librechat:auth, librechat:librechat:compose-pin, librechat:librechat:remote, librechat:librechat:local ### DATA-09: Professional secrecy Judgment rule: For professional secrecy holders (Berufsgeheimnisträger: lawyers, doctors, auditors, tax advisers) under § 203 StGB (DE) and Art. 321 StGB (CH), record whether the vendor signs a contractual secrecy undertaking or a § 203 agreement that binds its staff and its sub-processors; whether this comes as a dedicated offering or a paid add-on, and on which plan; who can access content in plain text (vendor staff, sub-processors, model providers); and the documentation provided. A DPA or a no-training clause alone is not a secrecy undertaking. Banking secrecy and FINMA outsourcing belong to DATA-06, not here. Record contract facts, not a legal assessment. #### Dust Verification status: Agent cross-checked (agent-cross-checked) Support: partial Public personnel/model handling statements exist; an explicit professional-secrecy contract and subprocessor flow-down remain unverified. Availability: generally-available Delivery: native Scope: surfaces=Dust web workspace, Dust EU web workspace; version=Public source pass 2026-10-21; no logged-in deployment observed; plan=Dust Business/Enterprise public policy; Enterprise and BYOK routes differ; add_ons=none stated; deployment_mode=multi-tenant-saas; region=EU/US storage and provider routes; no DE/CH-specific secrecy contract verified; os=not specified. Limitations: - No-training, ZDR marketing/policy assertions, EU hosting and a general privacy policy do not establish a professional-secrecy undertaking. - No legal suitability or compliance conclusion. - secrecy_undertaking — Does the vendor sign a secrecy undertaking or § 203 StGB / Art. 321 StGB agreement covering its own staff? [unknown]: The public security and platform privacy pages do not establish a signed §203 StGB/Art321 StGB professional-secrecy undertaking covering Dust staff. The Terms/Policies link did not expose readable contractual text in this pass. - subprocessor_coverage — Does the undertaking extend to its sub-processors, and is that documented? [unknown]: The privacy policy describes service-provider data-protection agreements and model no-training/retention conditions, but no dedicated professional-secrecy undertaking or its binding subprocessor flow-down was retrieved. - secrecy_offering — Is it a dedicated offering or add-on, and on which plan? [unknown]: No publicly documented German/Swiss professional-secrecy-specific plan or paid add-on was established. Enterprise single-tenancy/EU model options are not treated as such a contractual undertaking. - content_access — Who can access content in plain text: vendor staff, sub-processors, model providers? [partial]: The platform privacy policy limits personnel access to authorized staff and says customer context is processed by selected foundation-model providers; it asserts no provider disk retention/human-review logging. This vendor statement does not establish the complete customer-content plaintext access matrix, support countries, private contracts or every BYOK/MCP/tool route. Sources: dust:dust:privacy, dust:dust:training-policy, dust:dust:eu-models. - secrecy_documentation — Is the offering publicly documented (contract template, data-flow description, professional-body guidance)? [unknown]: Reviewed public security, privacy, EU-hosted model documentation and Enterprise plan scope plus targeted first-party secrecy search. Trust Center content was not available and Terms/Policies routed to unreadable external content. No public §203/Art321 template or professional-body guidance was retrieved. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: dust:dust:privacy, dust:dust:training-policy, dust:dust:eu-models #### LibreChat Verification status: Agent cross-checked (agent-cross-checked) Support: partial Customer-operated message storage and configurable local/external inference establish part of the content-access boundary. A LibreChat-scoped staff/sub-processor secrecy undertaking or professional-secrecy offering remains unverified. Availability: unknown Delivery: unknown Scope: surfaces=Customer-operated web application and server; version=Public documentation retrieved 2026-10-21; message schema and reference Compose pinned to v0.8.8 / e8f3be08623663d4ad7f7241e693c94469b63bb0; plan=MIT customer-operated LibreChat; optional commercial secrecy terms unverified. Public Demo and website policies checked only to distinguish their scope; add_ons=none stated; deployment_mode=on-premise; region=Customer-selected application, storage and model-service locations; no DE/CH professional-secrecy contract variant verified; os=not specified. Limitations: - The secrecy undertaking, sub-processor coverage, offering and professional-secrecy-specific documentation remain unknown. - content_access: Operator permissions, vendor/support grants, optional service access and provider staff access were not tested or contractually verified. - Technical data controls, a DPA, privacy wording and no-training statements do not establish the requested contractual undertaking. No legal compliance assessment is made. - secrecy_undertaking — Does the vendor sign a secrecy undertaking or § 203 StGB / Art. 321 StGB agreement covering its own staff? [unknown]: No LibreChat-scoped § 203 StGB / Art. 321 StGB undertaking binding vendor staff was obtained from the reviewed project documentation and public legal pages. The website privacy policy expressly excludes the self-hosted application; ClickHouse ownership does not establish a product-specific agreement. Sources: librechat:librechat:siteprivacy, librechat:librechat:owner. - subprocessor_coverage — Does the undertaking extend to its sub-processors, and is that documented? [unknown]: No self-hosted LibreChat professional-secrecy annex or documented extension to sub-processors was obtained. The separate Demo policy describes processor instruction/security obligations, which do not establish the undertaking requested here or coverage of a customer deployment. Sources: librechat:librechat:secrecy-demo-policy. - secrecy_offering — Is it a dedicated offering or add-on, and on which plan? [unknown]: No dedicated professional-secrecy edition, paid add-on, eligible plan or price was established in the public project/legal materials and targeted searches. The public Demo is evaluation-only and excludes confidential or regulated data; it is a separate surface. Sources: librechat:librechat:demo. - content_access — Who can access content in plain text: vendor staff, sub-processors, model providers? [partial]: The pinned reference deployment mounts customer-controlled MongoDB, uploads and logs; the message schema declares ordinary text/content fields. This places host/database operators inside the application trust boundary (an architectural inference, not a permissions audit). The first-chat and endpoint guides route prompts to a selected local or external model with customer credentials. Vendor/support access, optional service operators and provider human access remain deployment- and contract-specific and unverified. Sources: librechat:librechat:compose-pin, librechat:librechat:secrecy-message-pin, librechat:librechat:firstchat, librechat:librechat:custom. - secrecy_documentation — Is the offering publicly documented (contract template, data-flow description, professional-body guidance)? [unknown]: Public installation, endpoint and privacy documents describe technical data paths and distinguish website, Demo and self-hosted software. No LibreChat-specific DE/CH secrecy contract template, covered-party annex or professional-body guidance was located in the reviewed sources. Sources: librechat:librechat:firstchat, librechat:librechat:siteprivacy. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: librechat:librechat:siteprivacy, librechat:librechat:owner, librechat:librechat:secrecy-demo-policy, librechat:librechat:demo, librechat:librechat:compose-pin, librechat:librechat:secrecy-message-pin, librechat:librechat:firstchat, librechat:librechat:custom ## Commercials and vendor ### VEN-01: Pricing model and list prices Judgment rule: Record pricing unit (per seat, per active user, usage-based, platform fee, hybrid), list prices per plan in original currency with date, included usage and credits, add-on prices (connectors, agents, image generation) and overage rates. Cite public prices or dated quotes. Per-seat versus usage is a fact, not a quality judgment. #### Dust Verification status: Agent cross-checked (agent-cross-checked) Support: partial Business seat prices are public; only the documented prepaid route is described here, while conflicting Business overage eligibility remains unknown. Availability: generally-available Delivery: native Scope: surfaces=Dust web workspace, Dust EU web workspace; version=Public source pass 2026-10-21; no logged-in deployment observed; plan=Business Free/Pro/Max seat list pricing; Enterprise quoted pool and programmatic contracts are separate; add_ons=none stated; deployment_mode=multi-tenant-saas; region=US/global or EU workspace as named; EU-hosted-model restriction must be enabled separately; os=not specified. Limitations: - Documented findings only; unestablished atomic questions remain explicit gaps. - Do not present Business PAYG as established; retain list seat prices separately from the unresolved overage route. - pricing_unit — What is the pricing unit? [partial]: Business publishes per-seat subscriptions and included credits. Detailed Credit management describes prepaid Business top-ups and Enterprise PAYG; pricing FAQ separately describes enabled Business workspace overage, so Business post-allocation billing eligibility is unverified. Sources: dust:dust:pricing, dust:dust:credit-management. - list_prices — What are list prices per plan, with date and currency? [supported]: As retrieved 2026-10-21, ex-VAT Business Free is EUR 0; Pro EUR 30/month or EUR 24/month annually; Max EUR 150/month or EUR 120/month annually. Enterprise is quote-only. Sources: dust:dust:pricing. - included_usage — What usage is included? [supported]: Free has 500 lifetime credits, Pro 8,000/month and Max 40,000/month. Monthly credits do not roll over; Workspace seats draw from a contracted pool. Sources: dust:dust:credits, dust:dust:seats. - add_ons — Which features cost extra? [partial]: Business limits include three connectors, five remote MCP servers and five Spaces. Enterprise adds unlimited connector/MCP packaging, SCIM, audit, retention and support. Individual add-on tariffs are not published. Sources: dust:dust:pricing. - overage — What do usage beyond limits and premium models cost? [unknown]: Effective Business post-allocation billing is unknown: current pricing FAQ allows Pro/Max workspace overage, while Credit management reserves PAYG for Enterprise and describes prepaid Business top-ups. The USD0.01/credit Business programmatic rate does not resolve whether interactive overage is PAYG or prepaid. No account/contract test resolved this. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: dust:dust:pricing, dust:dust:credit-management, dust:dust:credits, dust:dust:seats #### LibreChat Verification status: Agent cross-checked (agent-cross-checked) Support: partial MIT software has no licence subscription; hosting, inference and optional runtime services are separately procured. Availability: generally-available Delivery: native Scope: surfaces=Customer-operated web application and server; version=Public documentation retrieved 2026-10-21; inspected source v0.8.8 / e8f3be08623663d4ad7f7241e693c94469b63bb0; plan=MIT customer-operated LibreChat; optional separately deployed services identified per answer; add_ons=none stated; deployment_mode=on-premise; region=Customer-selected infrastructure; model, OCR, search, observability and support services separately scoped; os=not specified. Limitations: - add_ons: Optional external code/OCR/search/image services need separately configured endpoints and credentials. Their rates depend on the operator’s contracts. - overage: Provider APIs charge their account holder; internal credit balances are governance configuration rather than a single LibreChat usage tariff. - pricing_unit — What is the pricing unit? [supported]: The pinned MIT permission grant has no per-user fee; self-hosting and model accounts are customer responsibilities. Sources: librechat:librechat:license, librechat:librechat:firstchat. - list_prices — What are list prices per plan, with date and currency? [supported]: The repository’s MIT licence permits use without a software fee as retrieved 2026-10-21; there is no bundled inference subscription in this scope. Sources: librechat:librechat:license, librechat:librechat:firstchat. - included_usage — What usage is included? [supported]: LibreChat ships with no model API key and no included provider token balance. Sources: librechat:librechat:firstchat. - add_ons — Which features cost extra? [partial]: Optional external code/OCR/search/image services need separately configured endpoints and credentials. Their rates depend on the operator’s contracts. Sources: librechat:librechat:code, librechat:librechat:ocr, librechat:librechat:web, librechat:librechat:images. - overage — What do usage beyond limits and premium models cost? [partial]: Provider APIs charge their account holder; internal credit balances are governance configuration rather than a single LibreChat usage tariff. Sources: librechat:librechat:balance, librechat:librechat:tokens, librechat:librechat:firstchat. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: librechat:librechat:license, librechat:librechat:firstchat, librechat:librechat:code, librechat:librechat:ocr, librechat:librechat:web, librechat:librechat:images, librechat:librechat:balance, librechat:librechat:tokens ### VEN-02: Scenario cost Judgment rule: Price three workloads for 12 months: 200, 1,000 and 5,000 users with a disclosed light/regular/power usage mix, in CHF and EUR excluding VAT. Include licences, platform fees, add-ons, model usage and, for self-hosted options, a stated operating assumption. Separate fixed, variable, included and unknown costs. No invented discounts. #### Dust Verification status: Agent cross-checked (agent-cross-checked) Support: partial ALM-14 shared cost assumptions v1.4, 2026-10-21 apply; all three target seat counts require unpriced Enterprise terms. Availability: generally-available Delivery: native Scope: surfaces=Dust web workspace, Dust EU web workspace; version=Public source pass 2026-10-21; no logged-in deployment observed; plan=Business credit-priced plan; Enterprise or legacy differences are named per answer; add_ons=none stated; deployment_mode=multi-tenant-saas; region=US/global or EU workspace as named; EU-hosted-model restriction must be enabled separately; os=not specified. Limitations: - Documented findings only; unestablished atomic questions remain explicit gaps. - cost_200 — Year-one cost at 200 users? [unknown]: Year-one cost at 200 users? For 200, 1,000 and 5,000 users over 12 months, the shared 60/30/10 message/token workload has no public applicable Enterprise seat quote or complete credit tariff. CHF/EUR ex-VAT totals remain unknown. ECB 2026-10-20 rates are recorded; the small-team Business price is not multiplied beyond its stated 100-person scope. - cost_1000 — Year-one cost at 1,000 users? [unknown]: Year-one cost at 1,000 users? For 200, 1,000 and 5,000 users over 12 months, the shared 60/30/10 message/token workload has no public applicable Enterprise seat quote or complete credit tariff. CHF/EUR ex-VAT totals remain unknown. ECB 2026-10-20 rates are recorded; the small-team Business price is not multiplied beyond its stated 100-person scope. - cost_5000 — Year-one cost at 5,000 users? [unknown]: Year-one cost at 5,000 users? For 200, 1,000 and 5,000 users over 12 months, the shared 60/30/10 message/token workload has no public applicable Enterprise seat quote or complete credit tariff. CHF/EUR ex-VAT totals remain unknown. ECB 2026-10-20 rates are recorded; the small-team Business price is not multiplied beyond its stated 100-person scope. - variable_share — Which part depends on usage? [supported]: Consumption adds token and action credits, including sub-agent work. Programmatic use and Enterprise PAYG draw from workspace pools separately from personal seat allocations. Sources: dust:dust:credits, dust:dust:credit-management. - unknown_parts — Which costs could not be priced? [supported]: Business is scoped to 100 people, so 200/1,000/5,000 seat Enterprise fees, committed pool and PAYG tariffs are unknown. The default Standard auto list is published, but a complete token-to-credit tariff is not, so no proxy replaces vendor-billed usage. Sources: dust:dust:pricing, dust:dust:models, dust:dust:credits. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: dust:dust:credits, dust:dust:credit-management, dust:dust:pricing, dust:dust:models, dust:dust:fx-ecb #### LibreChat Verification status: Agent cross-checked (agent-cross-checked) Support: partial ALM-14 shared cost assumptions v1.4, 2026-10-21: zero standard licence plus a disclosed Swiss infrastructure/operations and direct-BYO-model budget. Annual CHF/EUR figures are floors plus named unknown costs. Availability: generally-available Delivery: native Scope: surfaces=Customer-operated web application and server; version=Public documentation retrieved 2026-10-21; inspected source v0.8.8 / e8f3be08623663d4ad7f7241e693c94469b63bb0; plan=MIT customer-operated LibreChat; optional separately deployed services identified per answer; add_ons=Analyst-selected direct OpenAI API proxy, not a bundled model, Assumed operations staffing; optional commercial services excluded; deployment_mode=customer-cloud; region=Azure Switzerland North for the reference application infrastructure; standard OpenAI API inference has no Swiss/EU residency assumption; os=not specified. Limitations: - ALM-14 shared cost assumptions v1.4, 2026-10-21 are analyst inputs, not vendor sizing or an uptime promise. - The budget is a floor because named infrastructure and service costs are unpriced. - cost_200 — Year-one cost at 200 users? [partial]: ALM-14 shared cost assumptions v1.4, 2026-10-21. 200 users, 12 months excluding VAT: from CHF 34778.28 / EUR 37651.06 plus unpriced items. Includes zero standard-software licence, reference infrastructure EUR 4749.26, operations CHF 30000.00, and BYO model usage USD 475.20. This is an assumption-based budget floor, not a vendor quote or validated capacity plan. Sources: librechat:librechat:license, librechat:librechat:cost-vm, librechat:librechat:cost-pg, librechat:librechat:cost-fx, librechat:librechat:cost-model, librechat:librechat:cost-effort. - cost_1000 — Year-one cost at 1,000 users? [partial]: ALM-14 shared cost assumptions v1.4, 2026-10-21. 1,000 users, 12 months excluding VAT: from CHF 88600.40 / EUR 95919.02 plus unpriced items. Includes zero standard-software licence, reference infrastructure EUR 12605.23, operations CHF 75000.00, and BYO model usage USD 2376.00. This is an assumption-based budget floor, not a vendor quote or validated capacity plan. Sources: librechat:librechat:license, librechat:librechat:cost-vm, librechat:librechat:cost-pg, librechat:librechat:cost-fx, librechat:librechat:cost-model, librechat:librechat:cost-effort. - cost_5000 — Year-one cost at 5,000 users? [partial]: ALM-14 shared cost assumptions v1.4, 2026-10-21. 5,000 users, 12 months excluding VAT: from CHF 189541.24 / EUR 205197.83 plus unpriced items. Includes zero standard-software licence, reference infrastructure EUR 32214.49, operations CHF 150000.00, and BYO model usage USD 11880.00. This is an assumption-based budget floor, not a vendor quote or validated capacity plan. Sources: librechat:librechat:license, librechat:librechat:cost-vm, librechat:librechat:cost-pg, librechat:librechat:cost-fx, librechat:librechat:cost-model, librechat:librechat:cost-effort. - variable_share — Which part depends on usage? [partial]: Shared mix is 60% light at 100, 30% regular at 400 and 10% power at 1,500 messages/user/month, or 330 weighted messages. Each message is 3,000 input and 600 output tokens. Direct BYO proxy is gpt-6-luna, reasoning effort none, Standard short-context uncached rates USD 0.10/0.50 per million input/output tokens. It costs USD 2.376/user/year before optional tools, embedding, caching or regional processing. This proxy is an analyst assumption, not the product default. Sources: librechat:librechat:cost-model, librechat:librechat:cost-effort. - unknown_parts — Which costs could not be priced? [partial]: Unpriced additions are file/object storage beyond the assumed 500 GB PostgreSQL storage, VM OS/data disks, backup excess, egress/load balancers, Redis or additional database services, high availability, embeddings, OCR, search, audio/images, integration work and optional commercial services. MongoDB remains the primary store, so its infrastructure, file storage and other required dependencies are additional unpriced items; PostgreSQL is only the shared RAG/reference compute assumption. The shared VM/PG sizes have not been load-tested and do not replace each product’s documented runtime dependencies. Sources: librechat:librechat:cost-vm, librechat:librechat:cost-pg, librechat:librechat:license. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: librechat:librechat:license, librechat:librechat:cost-vm, librechat:librechat:cost-pg, librechat:librechat:cost-fx, librechat:librechat:cost-model, librechat:librechat:cost-effort ### VEN-03: Evaluation access Judgment rule: Distinguish free tier, self-serve trial, sales-led pilot, paid proof of concept and open-source self-install. Record card requirement, duration, seat limit, and which features (SSO, connectors, admin) the trial includes. Map which shared scenarios can be run without sales approval. #### Dust Verification status: Agent cross-checked (agent-cross-checked) Support: partial A limited Business free option enables self-serve evaluation; Enterprise features remain contract-scoped. Availability: generally-available Delivery: native Scope: surfaces=Dust web workspace, Dust EU web workspace; version=Public source pass 2026-10-21; no logged-in deployment observed; plan=Business credit-priced plan; Enterprise or legacy differences are named per answer; add_ons=none stated; deployment_mode=multi-tenant-saas; region=US/global or EU workspace as named; EU-hosted-model restriction must be enabled separately; os=not specified. Limitations: - Documented findings only; unestablished atomic questions remain explicit gaps. - free_tier — Is there a free tier for business use? [supported]: Business free supports up to five members with 500 lifetime credits per Free seat. Credits do not reset and a user cannot return to Free after upgrading. Sources: dust:dust:seats. - self_serve_trial — Is a self-serve business trial available? [partial]: The public site offers self-serve free workspace access, not a documented time-limited fully featured Enterprise trial. No card requirement was established. Sources: dust:dust:pricing, dust:dust:seats. - trial_scope — Do trials include SSO, connectors and admin features? [partial]: Business lists up to three connectors/five Spaces/five remote MCP servers and SSO at 5+ seats on demand. SCIM, audit and custom retention are Enterprise. Sources: dust:dust:pricing. - sales_gate — Does meaningful evaluation require sales contact? [supported]: Enterprise terms, pooled credits, advanced governance and support are sales-led; a small-team free evaluation is self-serve. Sources: dust:dust:pricing. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: dust:dust:seats, dust:dust:pricing #### LibreChat Verification status: Agent cross-checked (agent-cross-checked) Support: supported Customers can self-install without licence payment; the public demo is an evaluation-only alternative. Availability: generally-available Delivery: native Scope: surfaces=Customer-operated web application and server, LibreChat public evaluation demo; version=Public documentation retrieved 2026-10-21; inspected source v0.8.8 / e8f3be08623663d4ad7f7241e693c94469b63bb0; plan=MIT customer-operated LibreChat; optional separately deployed services identified per answer; add_ons=none stated; deployment_mode=on-premise; region=Customer-selected infrastructure; model, OCR, search, observability and support services separately scoped; os=not specified. - free_tier — Is there a free tier for business use? [supported]: MIT self-installation permits organizational use without a software licence fee. Sources: librechat:librechat:license. - self_serve_trial — Is a self-serve business trial available? [supported]: Self-installation has no trial clock or software card requirement; the free hosted Demo can be tried separately under its terms. Sources: librechat:librechat:local, librechat:librechat:demo. - trial_scope — Do trials include SSO, connectors and admin features? [supported]: Self-installed instances can configure SSO, Agents, file retrieval and administration with their own service accounts. It is not evidence that every experimental add-on is enabled by default. Sources: librechat:librechat:auth, librechat:librechat:agents, librechat:librechat:rag. - sales_gate — Does meaningful evaluation require sales contact? [supported]: The documented Docker install does not require sales approval; optional third-party services have separate account requirements. Sources: librechat:librechat:local, librechat:librechat:firstchat. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: librechat:librechat:license, librechat:librechat:local, librechat:librechat:demo, librechat:librechat:auth, librechat:librechat:agents, librechat:librechat:rag, librechat:librechat:firstchat ### VEN-04: Contract terms and pricing transparency Judgment rule: Record minimum seats or commitment, contract length, monthly versus annual billing, price-change rules, cancellation and downgrade terms, and invoicing in CHF/EUR. Classify pricing as transparent, partially specified, quote-only or unknown, naming the missing fields. Quote-only pricing is a disclosed constraint. #### Dust Verification status: Agent cross-checked (agent-cross-checked) Support: partial Business monthly and annual subscriptions are public; Enterprise seats are annual and prices contractual. Availability: generally-available Delivery: native Scope: surfaces=Dust web workspace, Dust EU web workspace; version=Public source pass 2026-10-21; no logged-in deployment observed; plan=Business credit-priced plan; Enterprise or legacy differences are named per answer; add_ons=none stated; deployment_mode=multi-tenant-saas; region=US/global or EU workspace as named; EU-hosted-model restriction must be enabled separately; os=not specified. Limitations: - Documented findings only; unestablished atomic questions remain explicit gaps. - minimum_commitment — What seat or spend minimum applies? [partial]: The free option has up to five members and Business targets up to 100 people; Enterprise numerical seat/spend minimum was not established. Sources: dust:dust:pricing, dust:dust:seats. - contract_length — What contract lengths are offered? [supported]: Business offers monthly and annual seats together; Enterprise seats are annual only. Sources: dust:dust:seats. - cancellation — How can the contract be cancelled or reduced? [supported]: Monthly seats unassign immediately with a prorated credit; annual seats return to stock for reassignment and are not refunded when someone leaves. Sources: dust:dust:seats. - local_invoicing — Can it be invoiced in CHF or EUR from a European entity? [partial]: Prices are displayed in EUR and the privacy policy identifies a French entity; CHF invoicing and the actual invoicing entity/currency were not contract-verified. Sources: dust:dust:pricing, dust:dust:privacy. - transparency_class — Is pricing transparent, partial, quote-only or unknown? [supported]: Partially specified: Business seat prices/credit allowances are public, while Enterprise pricing, complete usage conversion and negotiated terms require a quote. Sources: dust:dust:pricing, dust:dust:credits. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: dust:dust:pricing, dust:dust:seats, dust:dust:privacy, dust:dust:credits #### LibreChat Verification status: Agent cross-checked (agent-cross-checked) Support: partial The MIT licence has no seat or contract-term subscription; optional service contracts are outside that licence. Availability: generally-available Delivery: native Scope: surfaces=Customer-operated web application and server; version=Public documentation retrieved 2026-10-21; inspected source v0.8.8 / e8f3be08623663d4ad7f7241e693c94469b63bb0; plan=MIT customer-operated LibreChat; optional separately deployed services identified per answer; add_ons=none stated; deployment_mode=on-premise; region=Customer-selected infrastructure; model, OCR, search, observability and support services separately scoped; os=not specified. Limitations: - cancellation: An operator can stop running the application; this does not cancel its separately purchased cloud/model services or establish deletion of all copies. - local_invoicing: Can it be invoiced in CHF or EUR from a European entity? CHF/EUR commercial invoicing from a European entity was not established for optional project services. - transparency_class: Source licensing is transparent; a complete all-in hosted/support/inference tariff is not included in this self-hosted offer. - minimum_commitment — What seat or spend minimum applies? [supported]: The MIT permission grant has no user-count or spending minimum for software use. Sources: librechat:librechat:license. - contract_length — What contract lengths are offered? [supported]: There is no monthly/annual licence subscription in the MIT permission grant. External hosting/API contracts are separate. Sources: librechat:librechat:license. - cancellation — How can the contract be cancelled or reduced? [partial]: An operator can stop running the application; this does not cancel its separately purchased cloud/model services or establish deletion of all copies. Sources: librechat:librechat:license, librechat:librechat:local. - local_invoicing — Can it be invoiced in CHF or EUR from a European entity? [unknown]: local_invoicing: Can it be invoiced in CHF or EUR from a European entity? CHF/EUR commercial invoicing from a European entity was not established for optional project services. - transparency_class — Is pricing transparent, partial, quote-only or unknown? [partial]: Source licensing is transparent; a complete all-in hosted/support/inference tariff is not included in this self-hosted offer. Sources: librechat:librechat:license, librechat:librechat:firstchat. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: librechat:librechat:license, librechat:librechat:local, librechat:librechat:firstchat ### VEN-05: Source licence Judgment rule: Record licence per component (client, server, connectors, hosted extensions) and revision. Distinguish OSI open source, source-available with restrictions (branding clauses, user caps), and closed. Repository visibility is not a licence classification. Licences change between releases: record the current terms, and never carry a fork's historical licence over to the upstream product or the reverse. #### Dust Verification status: Agent cross-checked (agent-cross-checked) Support: partial The pinned public repository has an MIT root licence; hosted licensing and complete component coverage are separate. Availability: generally-available Delivery: native Scope: surfaces=Public dust-tt/dust repository root licence; version=dust-tt/dust commit2b1a4973af824f8ff970d585c5eb1f40ae76186c; root LICENSE only; plan=Public source component, not Business/Enterprise hosted-service rights; add_ons=none stated; deployment_mode=not specified; region=Repository licence has no deployment-region entitlement; os=not specified. Limitations: - MIT applies to the pinned repository root grant only; dependencies, supported self-host deployment, hosted terms and any closed Enterprise boundary remain unverified. - licence_type — Which licence applies to each component? [partial]: dust-tt/dust root LICENSE is MIT at commit 2b1a4973af824f8ff970d585c5eb1f40ae76186c. This does not classify every dependency or the contracted hosted service. Sources: dust:dust:licence. - branding_restrictions — Do licence terms restrict branding or user counts? [supported]: The MIT root grant requires preserving copyright and permission notices and does not itself impose branding or user-count caps. Sources: dust:dust:licence. - enterprise_features_closed — Are enterprise features in a separate closed edition? [unknown]: Are enterprise features in a separate closed edition? The pinned root LICENSE and README were retrieved; a complete nested-component licence inventory and separate closed Enterprise edition boundary were not established. Hosted terms were inaccessible. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: dust:dust:licence #### LibreChat Verification status: Agent cross-checked (agent-cross-checked) Support: partial The pinned main repository is MIT licensed; no user-count or product-branding clause appears in that licence. Availability: generally-available Delivery: native Scope: surfaces=Customer-operated web application and server; version=Public documentation retrieved 2026-10-21; inspected source v0.8.8 / e8f3be08623663d4ad7f7241e693c94469b63bb0; plan=MIT customer-operated LibreChat; optional separately deployed services identified per answer; add_ons=none stated; deployment_mode=on-premise; region=Customer-selected infrastructure; model, OCR, search, observability and support services separately scoped; os=not specified. Limitations: - enterprise_features_closed: The project documents a separate preview admin panel and optional code services. A closed enterprise-edition boundary for the main repository was not established. - licence_type — Which licence applies to each component? [supported]: The main client/server repository at v0.8.8 is MIT. Separate Code Interpreter, admin-panel, connector or hosted-service components need their own licence review. Sources: librechat:librechat:license. - branding_restrictions — Do licence terms restrict branding or user counts? [supported]: The MIT grant permits use/modification/distribution subject to retaining its copyright and permission notice; it includes no 50-user cap or mandatory UI branding clause. Sources: librechat:librechat:license. - enterprise_features_closed — Are enterprise features in a separate closed edition? [partial]: The project documents a separate preview admin panel and optional code services. A closed enterprise-edition boundary for the main repository was not established. Sources: librechat:librechat:admin, librechat:librechat:code, librechat:librechat:license. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: librechat:librechat:license, librechat:librechat:admin, librechat:librechat:code ### VEN-06: Data export and exit Judgment rule: Record admin export of chats, assistants, knowledge bases, prompts and audit logs, formats, export completeness, user self-export, and what survives contract end. Run an admin export on a trial. Exported chats alone do not make assistants and knowledge portable. #### Dust Verification status: Agent cross-checked (agent-cross-checked) Support: partial Business Conversation API and Enterprise Data Source API provide scoped data retrieval; agent export entitlement and complete tenant exit remain unknown. Availability: generally-available Delivery: native Scope: surfaces=Dust APIs and MCP server, Dust web workspace, Dust EU web workspace; version=Public source pass 2026-10-21; no logged-in deployment observed; plan=Business Conversation API and web analytics; Enterprise adds Data Source API. Agent-YAML endpoint entitlement is unverified.; add_ons=API keys with appropriate workspace/resource permissions; deployment_mode=multi-tenant-saas; region=US/global or EU workspace as named; EU-hosted-model restriction must be enabled separately; os=not specified. Limitations: - Conversation and document retrieval require the corresponding plan and resource permissions. Complete organization coverage, agent-export entitlement and the post-termination retrieval window remain unknown. - chat_export — Can admins export all chats? [partial]: Business pricing includes Conversation API; authorized API users retrieve accessible conversation messages with pagination. This does not establish one-click all-tenant admin export. Sources: dust:dust:get-conversation, dust:dust:pricing. - assistant_export — Can assistants and prompts be exported in a reusable format? [unknown]: The agent-YAML endpoint and public pricing were checked on 2026-10-21. Neither establishes its Business or Enterprise entitlement. Reusable assistant/prompt export for a named plan remains unknown. - knowledge_export — Can knowledge bases and their sources be exported? [partial]: Enterprise pricing adds Data Source API; the documented endpoint retrieves accessible data-source documents as JSON. Complete source/credential/permission portability and an all-tenant export are unverified. Sources: dust:dust:get-document, dust:dust:pricing. - export_format — Are formats documented and machine-readable? [partial]: Business Conversation API and Enterprise Data Source API return JSON for accessible resources. Web analytics separately exports CSV. These formats do not establish complete tenant export; no plan-scoped agent-YAML export is claimed. Sources: dust:dust:get-conversation, dust:dust:get-document, dust:dust:analytics, dust:dust:pricing. - post_termination_window — How long can data be retrieved after termination? [unknown]: How long can data be retrieved after termination? No trial admin exit/export was run; complete tenant coverage and the post-termination retrieval window were not established from the accessible API docs or unreadable contract route. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: dust:dust:get-conversation, dust:dust:pricing, dust:dust:get-document, dust:dust:analytics #### LibreChat Verification status: Agent cross-checked (agent-cross-checked) Support: partial Users can export conversations in machine-readable formats; organization-wide Agent/knowledge exit completeness remains unverified. Availability: generally-available Delivery: native Scope: surfaces=Customer-operated web application and server; version=Public documentation retrieved 2026-10-21; inspected source v0.8.8 / e8f3be08623663d4ad7f7241e693c94469b63bb0; plan=MIT customer-operated LibreChat; optional separately deployed services identified per answer; add_ons=none stated; deployment_mode=on-premise; region=Customer-selected infrastructure; model, OCR, search, observability and support services separately scoped; os=not specified. Limitations: - chat_export: Pinned client code exports a user-selected conversation, including optional branches/options. A documented admin bulk export of every user’s chats was not established. - assistant_export: Beta management APIs expose Agent/Skill objects, but a reusable complete assistant/prompt export fixture was not verified. - knowledge_export: Can knowledge bases and their sources be exported? No admin exit fixture, knowledge-source/index portability test or commercial post-termination retrieval window was verified; infrastructure backups are operator-owned. - post_termination_window: How long can data be retrieved after termination? No admin exit fixture, knowledge-source/index portability test or commercial post-termination retrieval window was verified; infrastructure backups are operator-owned. - chat_export — Can admins export all chats? [partial]: Pinned client code exports a user-selected conversation, including optional branches/options. A documented admin bulk export of every user’s chats was not established. Sources: librechat:librechat:export-pin. - assistant_export — Can assistants and prompts be exported in a reusable format? [partial]: Beta management APIs expose Agent/Skill objects, but a reusable complete assistant/prompt export fixture was not verified. Sources: librechat:librechat:api. - knowledge_export — Can knowledge bases and their sources be exported? [unknown]: knowledge_export: Can knowledge bases and their sources be exported? No admin exit fixture, knowledge-source/index portability test or commercial post-termination retrieval window was verified; infrastructure backups are operator-owned. - export_format — Are formats documented and machine-readable? [supported]: Conversation export code supports JSON, CSV, Markdown and text. This establishes format availability, not complete organization-data portability. Sources: librechat:librechat:export-pin. - post_termination_window — How long can data be retrieved after termination? [unknown]: post_termination_window: How long can data be retrieved after termination? No admin exit fixture, knowledge-source/index portability test or commercial post-termination retrieval window was verified; infrastructure backups are operator-owned. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: librechat:librechat:export-pin, librechat:librechat:api ### VEN-07: Support and service levels Judgment rule: Record support channels, languages (German, French, Italian), hours and time zone, local or regional presence, documented response times, uptime SLA and credits, public status page and incident history. Popularity is not a reliability score. #### Dust Verification status: Agent cross-checked (agent-cross-checked) Support: partial Business email support and Enterprise priority/SLA service are advertised; credit terms and local-language support remain open. Availability: generally-available Delivery: native Scope: surfaces=Dust web workspace, Dust EU web workspace; version=Public source pass 2026-10-21; no logged-in deployment observed; plan=Business credit-priced plan; Enterprise or legacy differences are named per answer; add_ons=none stated; deployment_mode=multi-tenant-saas; region=US/global or EU workspace as named; EU-hosted-model restriction must be enabled separately; os=not specified. Limitations: - Documented findings only; unestablished atomic questions remain explicit gaps. - support_channels — Which support channels are available on each plan? [supported]: Pricing lists Business email and Enterprise priority support with dedicated CSM; public support/community resources are available. Sources: dust:dust:pricing, dust:dust:support. - support_languages — Is support available in German, French or Italian? [unknown]: Is support available in German, French or Italian? Support/pricing/Enterprise pages and the official status URL were checked. German/French/Italian staffed support and detailed credited SLA terms were not established. - regional_presence — Is there a DACH or Swiss entity, team or partner? [partial]: The public policy identifies a Paris, France entity. A DACH/Swiss support team or named regional partner was not established. Sources: dust:dust:privacy. - uptime_sla — Is there an uptime SLA with credits? [partial]: Enterprise marketing claims 99.9% uptime and pricing includes an SLA; service-credit amounts, exclusions and claim rules were not retrieved. Sources: dust:dust:enterprise, dust:dust:pricing. - status_page — Is there a public status page with history? [partial]: Model-selection docs link https://dust.statuspage.io for incidents. Direct retrieval timed out, so incident history was not verified. Sources: dust:dust:models. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: dust:dust:pricing, dust:dust:support, dust:dust:privacy, dust:dust:enterprise, dust:dust:models #### LibreChat Verification status: Agent cross-checked (agent-cross-checked) Support: partial Public project/community channels are available; a binding business support SLA was not established. Availability: generally-available Delivery: native Scope: surfaces=Customer-operated web application and server; version=Public documentation retrieved 2026-10-21; inspected source v0.8.8 / e8f3be08623663d4ad7f7241e693c94469b63bb0; plan=MIT customer-operated LibreChat; optional separately deployed services identified per answer; add_ons=none stated; deployment_mode=on-premise; region=Customer-selected infrastructure; model, OCR, search, observability and support services separately scoped; os=not specified. Limitations: - support_channels: Public GitHub issue reporting and community channels are documented; a supported enterprise channel/response-time matrix was not established for MIT self-hosting. - support_languages: Is support available in German, French or Italian? No DE/FR/IT support coverage, CH/DACH legal/partner presence, priced uptime credits or product-service incident-history contract was verified in the public docs, site and release pass. - regional_presence: Is there a DACH or Swiss entity, team or partner? No DE/FR/IT support coverage, CH/DACH legal/partner presence, priced uptime credits or product-service incident-history contract was verified in the public docs, site and release pass. - uptime_sla: Is there an uptime SLA with credits? No DE/FR/IT support coverage, CH/DACH legal/partner presence, priced uptime credits or product-service incident-history contract was verified in the public docs, site and release pass. - status_page: Is there a public status page with history? No DE/FR/IT support coverage, CH/DACH legal/partner presence, priced uptime credits or product-service incident-history contract was verified in the public docs, site and release pass. - support_channels — Which support channels are available on each plan? [partial]: Public GitHub issue reporting and community channels are documented; a supported enterprise channel/response-time matrix was not established for MIT self-hosting. Sources: librechat:librechat:logging, librechat:librechat:firstchat. - support_languages — Is support available in German, French or Italian? [unknown]: support_languages: Is support available in German, French or Italian? No DE/FR/IT support coverage, CH/DACH legal/partner presence, priced uptime credits or product-service incident-history contract was verified in the public docs, site and release pass. - regional_presence — Is there a DACH or Swiss entity, team or partner? [unknown]: regional_presence: Is there a DACH or Swiss entity, team or partner? No DE/FR/IT support coverage, CH/DACH legal/partner presence, priced uptime credits or product-service incident-history contract was verified in the public docs, site and release pass. - uptime_sla — Is there an uptime SLA with credits? [unknown]: uptime_sla: Is there an uptime SLA with credits? No DE/FR/IT support coverage, CH/DACH legal/partner presence, priced uptime credits or product-service incident-history contract was verified in the public docs, site and release pass. - status_page — Is there a public status page with history? [unknown]: status_page: Is there a public status page with history? No DE/FR/IT support coverage, CH/DACH legal/partner presence, priced uptime credits or product-service incident-history contract was verified in the public docs, site and release pass. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: librechat:librechat:logging, librechat:librechat:firstchat ### VEN-08: Onboarding, training and adoption Judgment rule: Record vendor-led onboarding, admin training, end-user training (live, e-learning, languages), use-case libraries, champion programmes, change-management materials and their cost. Distinguish included services from paid professional services and partner offerings. #### Dust Verification status: Agent cross-checked (agent-cross-checked) Support: partial Dust has rollout and training material, with hands-on support packaged for Enterprise. Availability: generally-available Delivery: native Scope: surfaces=Dust web workspace, Dust EU web workspace; version=Public source pass 2026-10-21; no logged-in deployment observed; plan=Public learning material (no paid entitlement required to read); contracted CSM/onboarding is Enterprise; add_ons=none stated; deployment_mode=multi-tenant-saas; region=US/global or EU workspace as named; EU-hosted-model restriction must be enabled separately; os=not specified. Limitations: - Documented findings only; unestablished atomic questions remain explicit gaps. - onboarding_service — Is guided onboarding included? [supported]: Enterprise pricing includes dedicated CSM and onboarding; exact hours and scope are contractual. Sources: dust:dust:pricing. - user_training — Is end-user training offered, in which languages? [partial]: Public team onboarding and education resources are available. Course/live-session language coverage was not established. Sources: dust:dust:training, dust:dust:rollout. - use_case_library — Are role-based use cases and templates provided? [supported]: The public Use Cases and Guides index lists departmental guides for Sales, Support, Marketing, Engineering, Data Analytics, Knowledge, People, Product and Collaboration; the Sales guide supplies concrete agent examples and build links. This establishes public role-based guidance, not a separately verified in-product template entitlement. Sources: dust:dust:r2-department-guides, dust:dust:r2-sales-guide. - adoption_programme — Is there a champion or adoption programme? [partial]: The rollout guide covers launch, builder community, education and expansion; a contracted champion programme entitlement was not established. Sources: dust:dust:rollout. - service_cost — Which services cost extra? [partial]: Hands-on CSM/onboarding is Enterprise packaging; a numerical professional-services tariff was not established. Sources: dust:dust:pricing. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: dust:dust:pricing, dust:dust:training, dust:dust:rollout, dust:dust:r2-department-guides, dust:dust:r2-sales-guide #### LibreChat Verification status: Agent cross-checked (agent-cross-checked) Support: partial Self-hosting and Agent use have public setup guides; vendor-led training/adoption packages were not verified. Availability: generally-available Delivery: native Scope: surfaces=Customer-operated web application and server; version=Public documentation retrieved 2026-10-21; inspected source v0.8.8 / e8f3be08623663d4ad7f7241e693c94469b63bb0; plan=MIT customer-operated LibreChat; optional separately deployed services identified per answer; add_ons=none stated; deployment_mode=on-premise; region=Customer-selected infrastructure; model, OCR, search, observability and support services separately scoped; os=not specified. Limitations: - onboarding_service: Is guided onboarding included? No included guided rollout, local-language end-user training, champion programme, change-management service or professional-service price list was established from the official documentation and site. - user_training: Is end-user training offered, in which languages? No included guided rollout, local-language end-user training, champion programme, change-management service or professional-service price list was established from the official documentation and site. - use_case_library: The Agent and setup guides include examples and tutorials; a role-based enterprise training/use-case program was not established. - adoption_programme: Is there a champion or adoption programme? No included guided rollout, local-language end-user training, champion programme, change-management service or professional-service price list was established from the official documentation and site. - service_cost: Which services cost extra? No included guided rollout, local-language end-user training, champion programme, change-management service or professional-service price list was established from the official documentation and site. - onboarding_service — Is guided onboarding included? [unknown]: onboarding_service: Is guided onboarding included? No included guided rollout, local-language end-user training, champion programme, change-management service or professional-service price list was established from the official documentation and site. - user_training — Is end-user training offered, in which languages? [unknown]: user_training: Is end-user training offered, in which languages? No included guided rollout, local-language end-user training, champion programme, change-management service or professional-service price list was established from the official documentation and site. - use_case_library — Are role-based use cases and templates provided? [partial]: The Agent and setup guides include examples and tutorials; a role-based enterprise training/use-case program was not established. Sources: librechat:librechat:agents, librechat:librechat:firstchat. - adoption_programme — Is there a champion or adoption programme? [unknown]: adoption_programme: Is there a champion or adoption programme? No included guided rollout, local-language end-user training, champion programme, change-management service or professional-service price list was established from the official documentation and site. - service_cost — Which services cost extra? [unknown]: service_cost: Which services cost extra? No included guided rollout, local-language end-user training, champion programme, change-management service or professional-service price list was established from the official documentation and site. Checked 2026-10-21; agent cross-checked 2026-10-21; check again by 2026-12-20. Sources: librechat:librechat:agents, librechat:librechat:firstchat ## Sources ### dust:dust:models Auto models and the model picker. Users select a model per message or agent; auto choices use maintained ordered lists rather than task-based routing. - URL: https://docs.dust.tt/docs/user-documentation/agents/model-selection - Locator: Auto model lists; availability; API overrides - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:eu-models EU-hosted models. EU credit-plan workspaces using Dust model hosting can restrict agents and auto choices to EU-hosted models; BYOK is excluded from this setting. - URL: https://docs.dust.tt/docs/user-documentation/admins/admin-governance/eu-hosted-models - Locator: Availability; supported/excluded model table - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:changelog Dust changelog. Recent dated changes include German/French UI, Gemini 4 Argon and migration from File Generation to Computer. - URL: https://docs.dust.tt/docs/changelog - Locator: October 21, 16 and 14, 2026 entries - Evidence kind: documented - Source date: 2026-10-21 - Retrieved: 2026-10-21 ### dust:dust:files Conversation Files. Conversation uploads are scoped to one conversation and support documents, tabular data, images and audio. - URL: https://docs.dust.tt/docs/user-documentation/data-sources/conversation-files/conversation-files - Locator: Formats; size table; generating files - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:pricing Dust Pricing. Business supports up to 100 people with Free/Pro/Max seats; Enterprise pricing is sales-led and adds pooled credits, SCIM, audit, single tenancy and support. - URL: https://dust.tt/home/pricing - Locator: Business/Enterprise table and FAQ - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:custom Custom endpoints. Operators configure compatible endpoints and customer-held keys, including local and hosted services. - URL: https://www.librechat.ai/docs/quick_start/custom_endpoints - Locator: Endpoint setup; API key; base URL; model list - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:azure Azure OpenAI configuration. Azure OpenAI uses customer deployment groups, instance names and credentials. - URL: https://www.librechat.ai/docs/configuration/librechat_yaml/ai_endpoints/azure - Locator: Deployment groups; API credentials - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:bedrock AWS Bedrock object structure. Bedrock routes use customer AWS configuration and region selection. - URL: https://www.librechat.ai/docs/configuration/librechat_yaml/object_structure/aws_bedrock - Locator: Region; credentials; model selection - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:vertex Anthropic Vertex AI configuration. Claude can run through customer Google Vertex AI credentials and regions. - URL: https://www.librechat.ai/docs/configuration/librechat_yaml/object_structure/anthropic_vertex - Locator: Project and region configuration - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:modelspec Model specifications. Administrators can enforce configured model presets and defaults; available endpoints still depend on customer credentials. - URL: https://www.librechat.ai/docs/configuration/librechat_yaml/object_structure/model_specs - Locator: enforce; preset; modelSelect; addedEndpoints - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:agents Agents. Agents combine instructions, models, files, tools, skills and sharing permissions, with visible execution and configurable approval. - URL: https://www.librechat.ai/docs/features/agents - Locator: Builder; Capabilities; Version history; Sharing; Activity; tool approval - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:ocr OCR for documents. OCR can augment text extraction, or files can be sent to compatible provider APIs; format and model limitations remain. - URL: https://www.librechat.ai/docs/features/ocr - Locator: File Context; Upload as Text; Optional OCR; Limitations - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:releases LibreChat v0.8.8 release. The stable release documents agent controls, current model updates and explicitly experimental schedules/workspaces and beta Agents API. - URL: https://github.com/LibreChat-AI/LibreChat/releases/tag/v0.8.8 - Locator: Highlights, 2026-10-01 - Evidence kind: documented - Source date: 2026-10-01 - Retrieved: 2026-10-21 ### librechat:librechat:firstchat Your first chat. LibreChat ships without a model key; the public demo is for evaluation rather than private or durable data. - URL: https://www.librechat.ai/docs/quick_start/first_chat - Locator: Software you host yourself; endpoint credentials; public demo - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:code Code Interpreter API. A separate code execution service supports multiple languages and downloadable generated files; stateful/attached modes are experimental. - URL: https://www.librechat.ai/docs/features/code_interpreter - Locator: Setup; self-hosted service; workspace permissions; generated files - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:license LibreChat v0.8.8 MIT licence. The pinned repository is MIT licensed with copyright/permission notice preservation and an as-is disclaimer. - URL: https://raw.githubusercontent.com/LibreChat-AI/LibreChat/e8f3be08623663d4ad7f7241e693c94469b63bb0/LICENSE - Locator: LICENSE, permission grant and notice requirement - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:ragcfg RAG API configuration. OpenAI, Hugging Face and Ollama embedding configurations are documented; image choice affects local embedding support. - URL: https://www.librechat.ai/docs/configuration/rag_api - Locator: Embedding providers; lite image; full image - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:tiers Model access tiers. Admins restrict models and reasoning efforts at workspace, group or member level, with a separate published-agent bypass setting. - URL: https://docs.dust.tt/docs/user-documentation/admins/usage-seats-and-credits/model-access-tiers - Locator: Workspace, group and individual ceilings - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:spaces Spaces management. Spaces control knowledge and agent access; Business allows five and Enterprise 100 spaces. - URL: https://docs.dust.tt/docs/user-documentation/admins/spaces-management - Locator: Open/restricted spaces; limits; visibility - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:credits Credits. Consumption adds token and tool credits; a 600-AWU checkpoint can pause user-started root tasks, with exceptions. - URL: https://docs.dust.tt/docs/user-documentation/admins/usage-seats-and-credits/credits - Locator: Token/action credits; checkpoints - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:credit-management Credit management. Credit sources and block behavior vary by plan; beta Enterprise Pooled group budgets exclude programmatic consumption. - URL: https://docs.dust.tt/docs/user-documentation/admins/usage-seats-and-credits/credit-management - Locator: Personal caps; group budgets; programmatic use - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:access Granular access control. Roles, groups and resource ACLs govern features and sharing, with Entra groups synchronized at sign-in. - URL: https://www.librechat.ai/docs/features/access_control - Locator: Feature permissions; Resource ACLs; System grants; Groups - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:admin LibreChat Admin Panel. A separate browser management panel is available for testing and uses the underlying LibreChat admin APIs. - URL: https://www.librechat.ai/docs/features/admin_panel - Locator: Status Preview; separate service; admin APIs - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:balance Balance configuration. Optional token-credit balances initialize and refill user credits; tracking is disabled by default. - URL: https://www.librechat.ai/docs/configuration/librechat_yaml/object_structure/balance - Locator: enabled; startBalance; refill configuration - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:mod Automated moderation. Moderation scores abusive login/registration/message activity and may ban users; it is not a documented content DLP classifier. - URL: https://www.librechat.ai/docs/features/mod_system - Locator: Violation scoring and limits - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:balance-pin LibreChat balance admission implementation. The pinned middleware reserves user credits and throws a TOKEN_BALANCE error when the reservation cannot be admitted. - URL: https://raw.githubusercontent.com/LibreChat-AI/LibreChat/e8f3be08623663d4ad7f7241e693c94469b63bb0/packages/api/src/middleware/checkBalance.ts - Locator: packages/api/src/middleware/checkBalance.ts, checkBalance - Evidence kind: source-inspected - Source date: Not stated - Retrieved: 2026-10-21 - Commit: e8f3be08623663d4ad7f7241e693c94469b63bb0 - Path: packages/api/src/middleware/checkBalance.ts ### dust:dust:file-limits File size limits. The dedicated limits FAQ gives different image, audio and tabular caps than Conversation Files and excludes direct video uploads. - URL: https://docs.dust.tt/docs/user-documentation/getting-started/faq/file-size-limits - Locator: Upload caps; audio-only restriction - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:computer Computer. Computer executes code in temporary workspaces, handles Office/PDF files, performs documented OCR and restricts network access by allowlist. - URL: https://docs.dust.tt/docs/user-documentation/agents/tools/computer - Locator: File work; OCR; network control; secrets - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:drive Google Drive. Drive sync follows a dedicated connection account and selected content; scanned documents need conversation upload for OCR and Excel table queries need conversion to Sheets. - URL: https://docs.dust.tt/docs/user-documentation/admins/connections-management/google-drive - Locator: Connection identity; refresh; OCR; spreadsheet limitations - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:search Search data sources. Agents search selected knowledge sources and retrieve relevant document context. - URL: https://docs.dust.tt/docs/user-documentation/agents/knowledge/search-data-sources - Locator: Document selection; retrieval - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:rag RAG API. A separate RAG API indexes uploaded content in PostgreSQL/PGVector for retrieval. - URL: https://www.librechat.ai/docs/features/rag_api - Locator: Architecture; document indexing; embeddings; semantic search - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:text Upload files as text. Text upload injects extracted file content rather than relying only on vector retrieval. - URL: https://www.librechat.ai/docs/features/upload_as_text - Locator: Full text extraction and context - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:filecfg File configuration. File limits and formats are configurable by endpoint and file processing route. - URL: https://www.librechat.ai/docs/configuration/librechat_yaml/object_structure/file_config - Locator: File limits; MIME patterns; provider routes - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:r4-spreadsheet-parser LibreChat spreadsheet text parser. The built-in parser accepts XLSX/XLS/ODS and serializes each sheet as its name followed by CSV values. It does not establish preserved workbook layout, recalculated formulas or fixture accuracy. - URL: https://github.com/LibreChat-AI/LibreChat/blob/a378657583567310c841d652b0f6b00c83e2bd6b/packages/api/src/files/documents/crud.ts - Locator: getParserForMimeType; excelSheetToText, lines 109-129 - Evidence kind: source-inspected - Source date: 2026-10-24 - Retrieved: 2026-10-26 - Commit: a378657583567310c841d652b0f6b00c83e2bd6b - Path: packages/api/src/files/documents/crud.ts ### librechat:librechat:r4-text-current LibreChat current Upload as Text guide. The development guide lists XLSX/XLS/ODS parsing up to 15 MB and token-bounded text injection, not a formatted spreadsheet editor. - URL: https://www.librechat.ai/docs/features/upload_as_text - Locator: Built-in parser; supported documents; truncation - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-26 ### librechat:librechat:projects Projects. Projects persist files and instructions across a user’s chats but cannot be shared or transferred. - URL: https://www.librechat.ai/docs/features/projects - Locator: Availability; personal scope; files and instructions - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:permissions Access Controls and Permissions. Knowledge access is organized through admin-selected connection content and membership of open/restricted Spaces. - URL: https://docs.dust.tt/docs/user-documentation/admins/admin-governance/access-controls-and-permissions - Locator: Access policy; connection account; Spaces - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:microsoft Microsoft. Microsoft data is scoped to the connecting account and selected SharePoint sources; changes synchronize on a five-minute cycle. - URL: https://docs.dust.tt/docs/user-documentation/admins/connections-management/microsoft - Locator: Connection account; delegated and service-principal setup; sync - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:confluence Confluence. Confluence sync covers selected global spaces, excluding private spaces and pages with view restrictions. - URL: https://docs.dust.tt/docs/user-documentation/admins/connections-management/confluence - Locator: Scope; restricted pages; refresh - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:jira Beta: Import Jira issues. A separately documented beta route imports Jira issues as a custom connection. - URL: https://docs.dust.tt/docs/user-documentation/data-sources/custom-connections/beta-import-jira-issues - Locator: Import path - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:sources What data sources can be connected to Dust?. Official guides identify managed connections and custom API/import routes. - URL: https://docs.dust.tt/docs/user-documentation/getting-started/faq/general-questions/what-data-sources-can-be-connected-to-dust - Locator: Named sources - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:tools Adding an MCP Server. Admins add remote MCP servers with OAuth or bearer credentials and can restrict tools to skills. - URL: https://docs.dust.tt/docs/user-documentation/admins/tools-management/adding-an-mcp-server - Locator: Remote/custom servers; auth; tool restriction - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:r2-bigquery Dust BigQuery connection. Dust stores BigQuery metadata and runs table queries through a dedicated service account; dataset content is not synchronized into the index. - URL: https://docs.dust.tt/docs/user-documentation/admins/connections-management/bigquery.md - Locator: No content sync; service-account setup, one location and Space table selection - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:sharepoint SharePoint and OneDrive integration. The Microsoft file picker uses Entra token reuse and delegated on-behalf-of downloads to attach files. - URL: https://www.librechat.ai/docs/configuration/sharepoint - Locator: Prerequisites; delegated permissions; admin consent - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:mcp MCP. MCP servers can be configured through YAML or the permission-gated UI, including per-user OAuth. - URL: https://www.librechat.ai/docs/features/mcp - Locator: YAML/UI setup; authentication; ownership and selection - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:mcpcfg MCP server configuration. Configuration includes stdio, SSE and Streamable HTTP transports and server-specific credentials. - URL: https://www.librechat.ai/docs/configuration/librechat_yaml/object_structure/mcp_servers - Locator: Transport examples; headers; OAuth - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:credentials Personal vs Shared Credentials for Tools & MCP Servers. Tools can use a shared account or each user OAuth identity; even personal auth needs initial admin setup. - URL: https://docs.dust.tt/docs/user-documentation/admins/tools-management/personal-vs-shared-credentials - Locator: Shared versus personal authentication - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:security Dust Security. Dust claims EU/US hosting, AES-256/TLS, no model training, provider zero retention and SOC 2 Type II. - URL: https://dust.tt/home/security - Locator: Hosting, training, encryption and certification statements - Evidence kind: vendor-claimed - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:compose-pin LibreChat Docker service configuration. The pinned example runs a MongoDB primary store and a separate PGVector/RAG service with customer-mounted data. - URL: https://raw.githubusercontent.com/LibreChat-AI/LibreChat/e8f3be08623663d4ad7f7241e693c94469b63bb0/docker-compose.yml - Locator: docker-compose.yml, api/mongodb/vectordb/rag_api services - Evidence kind: source-inspected - Source date: Not stated - Retrieved: 2026-10-21 - Commit: e8f3be08623663d4ad7f7241e693c94469b63bb0 - Path: docker-compose.yml ### dust:dust:deep Deep-dive Agent. Deep-dive plans, delegates and writes cited reports, using Company Data knowledge and selected tools; Table Query is excluded. - URL: https://docs.dust.tt/docs/user-documentation/agents/default-agents/deep-dive-agent - Locator: Research flow; citations; limitations - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:memory Agent Memory. Enabled agents retain user-private memory across conversations; users can view and delete it and it persists until deletion. - URL: https://docs.dust.tt/docs/user-documentation/agents/tools/agent-memory - Locator: Adding the tool; viewing/deleting; privacy - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:builder Create your first agent. Agents combine instructions, models, knowledge and tools with controlled sharing. - URL: https://docs.dust.tt/docs/user-documentation/agents/create-your-first-agent - Locator: Instructions; tools; sharing - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:memory User memory. Configured user memory stores cross-chat key/value context with user editing controls; it does not index all past chats. - URL: https://www.librechat.ai/docs/features/memory - Locator: Key/value store; controls; disabled by default - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:governance Workspace Governance: Roles, Groups & Permissions. Admins delegate manager/group permissions for agents, skills, models, billing and security; grants are additive. - URL: https://docs.dust.tt/docs/user-documentation/admins/admin-governance/workspace-governance-roles-groups-and-permissions - Locator: Roles; permission matrix; additive access - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:images Image generation and editing. Agents can use configured image tools for generation/editing; provider credentials and model parameters are separate. - URL: https://www.librechat.ai/docs/features/image_gen - Locator: OpenAI Image Tools; provider setup; admin configuration - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:web Web Search & Browse. The tool runs Google queries and browses results, with agent-level configuration and format options. - URL: https://docs.dust.tt/docs/user-documentation/agents/tools/web-search-and-browse - Locator: Search and browsing flow - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:web Web search. Search can use Serper, SearXNG, Tavily or Keenable, with independently configured scraping and optional reranking. - URL: https://www.librechat.ai/docs/features/web_search - Locator: Provider, scraper and reranker configuration - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:triggers Triggers. Agents can run automatically from schedule or event triggers. - URL: https://docs.dust.tt/docs/user-documentation/agents/triggers/index - Locator: Schedules and events - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:schedules Schedules. Agents can run on recurring schedules. - URL: https://docs.dust.tt/docs/user-documentation/agents/triggers/schedules - Locator: Schedule setup - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:webhooks Webhooks. Admins configure webhook sources; members attach filtered triggers to agents, with optional signature verification for custom sources. - URL: https://docs.dust.tt/docs/user-documentation/agents/triggers/webhooks/index - Locator: Built-in and custom event sources - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:schedule Scheduled Chats. Opt-in experimental schedules execute saved Agents with cron/time zones and recheck access and balance before running. - URL: https://www.librechat.ai/docs/features/scheduled_chats - Locator: Experimental and disabled by default; runtime and permissions - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:api Agents API beta. The beta API exposes chat/Responses inference, Agent/Skill management and authenticated event delivery. - URL: https://www.librechat.ai/docs/features/agents_api - Locator: Inference; management; Agent Events; authentication - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:pods Pods overview. Pods combine shared conversations, tasks and files; open Pods are workspace-browsable and restricted Pods require invitation. - URL: https://docs.dust.tt/docs/user-documentation/pods/overview - Locator: Shared workspace and visibility - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:api-keys API keys. Dust API keys have read/admin scopes, data access and enforceable per-key monthly caps. - URL: https://docs.dust.tt/docs/user-documentation/admins/api-keys - Locator: Scopes; Space access; caps - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:mcp-server Dust MCP Server. Dust exposes workspace knowledge, agents, conversations, files and Pods to OAuth MCP clients under the authenticated user permissions. - URL: https://docs.dust.tt/docs/user-documentation/agents/integrations/dust-mcp-server - Locator: Region URLs; OAuth; user permissions - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:upsert Upsert a document in a data source. The API loads or updates documents in a scoped data source. - URL: https://docs.dust.tt/api-reference/datasources/upsert-a-document-in-a-data-source - Locator: Endpoint and authentication - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:get-conversation Get a conversation. The API retrieves accessible conversation messages with pagination. - URL: https://docs.dust.tt/api-reference/conversations/get-a-conversation - Locator: Response schema - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:browser Browser Extension. An official extension supports Chrome, Firefox and Chromium-based browsers. - URL: https://docs.dust.tt/docs/user-documentation/agents/integrations/browser-extension - Locator: Browsers and capabilities - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:retention Temporary chat and data retention. Operators configure expiry for chats/messages/files/shared links; normal chats do not expire by default and memories are excluded. - URL: https://www.librechat.ai/docs/features/temporary_chat - Locator: Default 30-day temporary retention; all/ephemeral; deadlines; memories - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:insights Admin Insights. Insights reports persisted MongoDB activity and identifiable recent conversations; it is not monetary billing analytics. - URL: https://www.librechat.ai/docs/features/insights - Locator: Opt-in; capabilities; metrics; tenant scope - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:local Docker installation. Docker self-installation runs LibreChat with its configured service dependencies. - URL: https://www.librechat.ai/docs/local/docker - Locator: Installation; configuration; updating - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:teams Dust in Teams. An official Teams integration runs Dust agents within Microsoft Teams after admin setup. - URL: https://docs.dust.tt/docs/user-documentation/agents/integrations/dust-in-teams - Locator: Setup and usage - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:images Image Generation. Agents generate images from text or references, up to 14 reference images, with SynthID and content restrictions. - URL: https://docs.dust.tt/docs/user-documentation/agents/tools/image-generation - Locator: Reference images; parallel generation; safety - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:speech Voice and sound generation. The speech tool generates MP3 narration or multi-speaker dialogue with selectable languages and styles. - URL: https://docs.dust.tt/docs/user-documentation/agents/tools/voice-and-sound-generation - Locator: Speech Generator - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:meetings Meeting transcripts. Dust consumes transcripts from connected notetaker and sales tools; Gong private transcripts are excluded. - URL: https://docs.dust.tt/docs/user-documentation/agents/integrations/meeting-transcripts - Locator: Providers and scope - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:speech Speech configuration. Administrators configure audio provider engines and models separately from text generation. - URL: https://www.librechat.ai/docs/configuration/librechat_yaml/object_structure/speech - Locator: Speech-to-text and text-to-speech engines - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:speech-conversation-r2 LibreChat conversation-mode speech settings. Conversation mode chains transcription, auto-send and TTS playback for hands-free turns after microphone activation; engines and audio routing are separately configured. - URL: https://www.librechat.ai/docs/configuration/stt_tts - Locator: Speech Tab: conversationMode, autoTranscribeAudio, autoSendText, automaticPlayback; engine and allowedAddresses settings - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:frames Frames overview. Frames provide persistent interactive agent output that can be refined and shared. - URL: https://docs.dust.tt/docs/user-documentation/agents/frames/overview - Locator: Creating and editing interactive content - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:artifacts Artifacts. Artifacts render generated React, HTML and Mermaid content in the chat interface. - URL: https://www.librechat.ai/docs/features/artifacts - Locator: React, HTML and Mermaid previews - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:pod-chats Pod conversations. All Pod members can read and contribute to Pod conversations, which are automatically indexed. - URL: https://docs.dust.tt/docs/user-documentation/pods/conversations - Locator: Visibility; indexing; participation - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:share Shareable links. Shared links publish a transcript branch and optional file snapshots independently of the owner’s live file ACL. - URL: https://www.librechat.ai/docs/features/shareable_links - Locator: Authentication defaults; file snapshots; continuation - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:i18n-pin LibreChat locale registration. The pinned client registers English, German, French and Italian translations among its supported locales. - URL: https://raw.githubusercontent.com/LibreChat-AI/LibreChat/e8f3be08623663d4ad7f7241e693c94469b63bb0/client/src/locales/i18n.ts - Locator: client/src/locales/i18n.ts, language resources - Evidence kind: source-inspected - Source date: Not stated - Retrieved: 2026-10-21 - Commit: e8f3be08623663d4ad7f7241e693c94469b63bb0 - Path: client/src/locales/i18n.ts ### dust:dust:sso Single Sign-On. Admins configure and enforce SAML SSO, which disables other social sign-in paths for the workspace. - URL: https://docs.dust.tt/docs/user-documentation/admins/admin-governance/single-sign-on-sso/single-sign-on-sso - Locator: Setup; enforcement - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:enterprise Dust for Enterprise. The Enterprise marketing page claims 365-day audit retention, 99.9% uptime and support for 10,000+ users. - URL: https://dust.tt/home/enterprise - Locator: Security and scale - Evidence kind: vendor-claimed - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:auth Authentication system. Local/LDAP accounts can be required to use TOTP; federated MFA belongs to the identity provider and account creation/deletion scripts are documented. - URL: https://www.librechat.ai/docs/configuration/authentication - Locator: Required two-factor authentication; Registration; User scripts - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:openid OAuth2 and OIDC. LibreChat documents OAuth2/OIDC providers including Microsoft Entra, Google, Keycloak and Auth0. - URL: https://www.librechat.ai/docs/configuration/authentication/OAuth2-OIDC - Locator: Supported providers; setup - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:saml SAML authentication. SAML is a documented authentication option for the self-hosted application. - URL: https://www.librechat.ai/docs/configuration/authentication/SAML - Locator: Overview and requirements - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:entra Azure Entra authentication. Entra OIDC supports group-based admission and Graph integration using configured scopes and token reuse. - URL: https://www.librechat.ai/docs/configuration/authentication/OAuth2-OIDC/azure - Locator: Group claims; token reuse; Microsoft Graph integration - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:provision Users and groups provisioning. Enterprise SCIM creates, updates and deactivates users and syncs groups; groups can grant Admin/Manager roles. - URL: https://docs.dust.tt/docs/user-documentation/admins/admin-governance/users-and-groups-provisioning - Locator: SCIM 2.0; group role mapping - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:seats Seat management. Admins assign independent roles and Free/Pro/Max seats; annual stock and already-allocated credits have specific reclaim rules. - URL: https://docs.dust.tt/docs/user-documentation/admins/usage-seats-and-credits/seat-management - Locator: Seat assignments; billing cycles; leaving - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:analytics Analytics. Consumption reports break down users, groups, models, tools, agents and Pods; exports omit message content and can double-attribute group usage. - URL: https://docs.dust.tt/analytics - Locator: Attribution; CSV; content exclusion - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:tokens Token usage. Usage and estimated costs use recorded tokens and configured provider rates; transactions are distinct from balance enforcement. - URL: https://www.librechat.ai/docs/configuration/token_usage - Locator: Context cost; transactions; per-endpoint token rates - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:audit Audit Logs. Enterprise logs distinguish users, API keys and system actors with CSV and continuous SIEM delivery. - URL: https://docs.dust.tt/docs/user-documentation/admins/audit-logs/audit-logs - Locator: Enterprise scope; actor types; SIEM - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:logging Logging. Backend operational logs are written to customer files and can feed external tools; the guide describes 14-day rotation rather than a compliance archive. - URL: https://www.librechat.ai/docs/configuration/logging - Locator: Error/debug logs; rotation; external tools - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:r4-content-filter LibreChat development content-filter configuration. Opt-in base-YAML policies inspect selected message/file content; block and audit differ from limited Agent text redaction. This is development documentation, not a hosted DLP assurance. - URL: https://www.librechat.ai/docs/configuration/librechat_yaml/object_structure/message_filter - Locator: Source-Aware Filters; Pattern Configuration; Redaction; File Fields - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-26 ### librechat:librechat:r4-pii-runtime LibreChat content-inspection runtime. Source-aware configured inspection executes in the LibreChat API. Missing policy disables that scope; configured actions and fields bound inspection. - URL: https://github.com/LibreChat-AI/LibreChat/blob/a378657583567310c841d652b0f6b00c83e2bd6b/packages/api/src/protection/runtime.ts - Locator: createConfiguredContentInspector; source rule configuration - Evidence kind: source-inspected - Source date: 2026-10-24 - Retrieved: 2026-10-26 - Commit: a378657583567310c841d652b0f6b00c83e2bd6b - Path: packages/api/src/protection/runtime.ts ### librechat:librechat:r4-patterns LibreChat pattern detector. Default starter patterns detect credential-shaped tokens/headers. Operator custom patterns use bounded regular expressions; a generic comprehensive PII classifier is not established. - URL: https://github.com/LibreChat-AI/LibreChat/blob/a378657583567310c841d652b0f6b00c83e2bd6b/packages/api/src/protection/detectors/pattern.ts - Locator: STARTER_PATTERNS; configured RE2JS rules - Evidence kind: source-inspected - Source date: 2026-10-24 - Retrieved: 2026-10-26 - Commit: a378657583567310c841d652b0f6b00c83e2bd6b - Path: packages/api/src/protection/detectors/pattern.ts ### librechat:librechat:r4-private-submission LibreChat private text submission. The private text ingress permits configured redaction only on eligible fresh interactive Agent text. Original content is encrypted; edit/resume/file paths do not gain blanket redaction. - URL: https://github.com/LibreChat-AI/LibreChat/blob/a378657583567310c841d652b0f6b00c83e2bd6b/packages/api/src/protection/private/submission.ts - Locator: createPrivateTextIngress; fresh Agent text and clientRequestId checks - Evidence kind: source-inspected - Source date: 2026-10-24 - Retrieved: 2026-10-26 - Commit: a378657583567310c841d652b0f6b00c83e2bd6b - Path: packages/api/src/protection/private/submission.ts ### librechat:librechat:r4-filter-middleware LibreChat API content-filter middleware. The server middleware applies configured inspection and rejects blocked/uninspectable requests with HTTP 400. No deployment or detection fixture was run. - URL: https://github.com/LibreChat-AI/LibreChat/blob/a378657583567310c841d652b0f6b00c83e2bd6b/packages/api/src/middleware/messageFilterPii.ts - Locator: createMessageFilterPii; contentFilterBlockResponse; uninspectable response - Evidence kind: source-inspected - Source date: 2026-10-24 - Retrieved: 2026-10-26 - Commit: a378657583567310c841d652b0f6b00c83e2bd6b - Path: packages/api/src/middleware/messageFilterPii.ts ### librechat:librechat:r4-config-1318 LibreChat config v1.3.18 changelog. Dated development config changes add limited Agent text redaction and ephemeral retention mode. These are not a promise of universal redaction or a vendor zero-retention contract. - URL: https://www.librechat.ai/changelog/config_v1.3.18 - Locator: October 22, 2026; opt-in PII redaction; retention mode - Evidence kind: documented - Source date: 2026-10-22 - Retrieved: 2026-10-26 ### librechat:librechat:interface Interface configuration. Administrators configure UI feature permissions and usage terms in the self-hosted deployment. - URL: https://www.librechat.ai/docs/configuration/librechat_yaml/object_structure/interface - Locator: Feature permissions; custom welcome and terms - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:privacy Dust Platform Privacy Policy. The policy identifies Permutation Labs in Paris and distinguishes its controller processing from customer processor processing, with non-EEA transfer safeguards. - URL: https://dust.tt/home/platform-privacy - Locator: About Us; sections 2, 6 and 7 - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:remote Remote hosting. The application can be deployed on customer-selected remote servers with operator-managed security. - URL: https://www.librechat.ai/docs/remote - Locator: Hosting options; security considerations - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:demo LibreChat Demo terms. The free hosted Demo is offered as-is with possible data cleanups and no durable business-service guarantee. - URL: https://www.librechat.ai/demo/terms - Locator: Evaluation scope; retention; disclaimer - Evidence kind: documented - Source date: 2026-03-12 - Retrieved: 2026-10-21 ### dust:dust:training-policy Dust Platform Privacy Policy. The policy asserts listed providers do not train, retain or log Customer Content for human review; this source pass did not retrieve the customer DPA. - URL: https://dust.tt/home/platform-privacy - Locator: Section 6, Foundational Model Providers - Evidence kind: vendor-claimed - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:delete-chat Can I delete or rename a conversation?. Users can delete or rename their conversations. - URL: https://docs.dust.tt/docs/user-documentation/getting-started/faq/managing-agents/can-i-delete-or-rename-a-conversation - Locator: Conversation actions - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:settings Personal settings. Users configure interface language, audio and personal data controls. - URL: https://www.librechat.ai/docs/features/settings - Locator: Language; speech; data controls - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:owner ClickHouse acquisition of LibreChat. ClickHouse announced the acquisition of LibreChat and that Danny Avila and the team joined ClickHouse. - URL: https://clickhouse.com/blog/clickhouse-acquires-librechat - Locator: Announcement, November 4, 2025 - Evidence kind: documented - Source date: 2025-11-04 - Retrieved: 2026-10-21 ### librechat:librechat:siteprivacy LibreChat website privacy policy. Website, demo and self-hosted software are distinct privacy scopes; website terms do not establish a customer deployment DPA. - URL: https://www.librechat.ai/privacy - Locator: Scope and exclusions - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:secrecy-message-pin LibreChat v0.8.8 message storage schema. The pinned Mongoose schema declares message text as a String and content as mixed fields; this source does not establish operator exclusion from live content. - URL: https://raw.githubusercontent.com/LibreChat-AI/LibreChat/e8f3be08623663d4ad7f7241e693c94469b63bb0/packages/data-schemas/src/schema/message.ts - Locator: packages/data-schemas/src/schema/message.ts, text at lines 55-58 and content at lines 139-143 - Evidence kind: source-inspected - Source date: Not stated - Retrieved: 2026-10-21 - Commit: e8f3be08623663d4ad7f7241e693c94469b63bb0 - Path: packages/data-schemas/src/schema/message.ts ### librechat:librechat:secrecy-demo-policy LibreChat Demo privacy policy. The Demo policy describes model-provider transmission and contractual processor instruction/security duties for the Demo; it does not cover a customer self-hosted deployment. - URL: https://www.librechat.ai/demo/privacy - Locator: Scope; section 4 Third-Party AI Model Providers; section 6 Data Sharing and Sub-Processors - Evidence kind: documented - Source date: 2026-03-12 - Retrieved: 2026-10-21 ### dust:dust:fx-ecb ECB historical euro reference rates. EUR 1 = USD 1.1215 = CHF 0.9237 on October 20, 2026. - URL: https://www.ecb.europa.eu/stats/eurofxref/eurofxref-hist.xml - Locator: Cube time=2026-10-20, USD and CHF - Evidence kind: documented - Source date: 2026-10-20 - Retrieved: 2026-10-21 ### librechat:librechat:cost-vm Azure Retail Prices, Switzerland North Linux VMs. EUR hourly PAYG rates retrieved 2026-10-21 are 0.2226, 0.4453 and 0.8905 for D4s, D8s and D16s v5 respectively. - URL: https://prices.azure.com/api/retail/prices?currencyCode=EUR&%24filter=armRegionName+eq+%27switzerlandnorth%27+and+serviceName+eq+%27Virtual+Machines%27+and+%28armSkuName+eq+%27Standard_D4s_v5%27+or+armSkuName+eq+%27Standard_D8s_v5%27+or+armSkuName+eq+%27Standard_D16s_v5%27%29+and+priceType+eq+%27Consumption%27 - Locator: Consumption; Virtual Machines Dsv5 Series; Standard_D4s_v5/D8s_v5/D16s_v5 - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:cost-pg Azure Retail Prices, PostgreSQL Flexible Server. EUR hourly PAYG GP compute is 0.2288, 0.4576 and 0.9152 for 2/4/8 vCore; storage is EUR 0.1325 per GB-month. - URL: https://prices.azure.com/api/retail/prices?currencyCode=EUR&%24filter=armRegionName+eq+%27switzerlandnorth%27+and+serviceName+eq+%27Azure+Database+for+PostgreSQL%27+and+priceType+eq+%27Consumption%27 - Locator: General Purpose Ddsv5 Series Compute, 2/4/8 vCore; Flex Server Storage - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:cost-fx ECB euro reference rate history. The 2026-10-20 reference rates are EUR 1 = USD 1.1215 and CHF 0.9237. - URL: https://www.ecb.europa.eu/stats/eurofxref/eurofxref-hist.xml - Locator: Cube time=2026-10-20; USD and CHF - Evidence kind: documented - Source date: 2026-10-20 - Retrieved: 2026-10-21 ### librechat:librechat:cost-model OpenAI API pricing. Standard gpt-6-luna API price is USD 0.10 per million input tokens and USD 0.50 per million output tokens; caching, batch and regional premiums are separate. - URL: https://developers.openai.com/api/docs/pricing - Locator: Standard, short-context gpt-6-luna row - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:cost-effort GPT-6 Luna model documentation. The model accepts reasoning effort none, the lowest documented effort, and lists the same standard text-token rates. - URL: https://developers.openai.com/api/docs/models/gpt-6-luna - Locator: Reasoning effort and text-token pricing - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:licence Dust repository LICENSE at pinned commit. The pinned public repository root licence is MIT; hosted service terms and third-party components are separate. - URL: https://github.com/dust-tt/dust/blob/2b1a4973af824f8ff970d585c5eb1f40ae76186c/LICENSE - Locator: Root LICENSE, MIT permission grant - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:get-document Retrieve a document from a data source. The API retrieves a source document; this does not establish a complete one-click tenant export. - URL: https://docs.dust.tt/api-reference/datasources/retrieve-a-document-from-a-data-source - Locator: Endpoint and response - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### librechat:librechat:export-pin LibreChat conversation export implementation. The pinned user conversation exporter supports JSON, CSV, Markdown and text, plus screenshot export. - URL: https://raw.githubusercontent.com/LibreChat-AI/LibreChat/e8f3be08623663d4ad7f7241e693c94469b63bb0/client/src/hooks/Conversations/useExportConversation.ts - Locator: client/src/hooks/Conversations/useExportConversation.ts, export dispatch - Evidence kind: source-inspected - Source date: Not stated - Retrieved: 2026-10-21 - Commit: e8f3be08623663d4ad7f7241e693c94469b63bb0 - Path: client/src/hooks/Conversations/useExportConversation.ts ### dust:dust:support Dust Support. The vendor provides public support entry points and community/documentation resources. - URL: https://dust.tt/home/support - Locator: Support channels - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:rollout Dust Rollout Guide. Public rollout guidance covers setup, launch, team education and measurement. - URL: https://docs.dust.tt/docs/user-documentation/getting-started/dust-rollout-guide/index - Locator: Rollout phases - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:training Team onboarding material. Dust provides reusable team-onboarding material. - URL: https://docs.dust.tt/docs/user-documentation/getting-started/dust-rollout-guide/team-onboarding-material - Locator: Training resources - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:r2-department-guides Dust Use Cases and Guides. Public guides are organized by Sales, Support, Marketing, Engineering, Data, Knowledge, People, Product and Collaboration. - URL: https://docs.dust.tt/docs/user-documentation/getting-started/use-cases-and-guides/index.md - Locator: Use Cases by Department - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21 ### dust:dust:r2-sales-guide Dust Sales use-case guide. The Sales guide provides concrete role-based agent examples; this is public learning material, not a contracted training entitlement. - URL: https://docs.dust.tt/docs/user-documentation/getting-started/use-cases-and-guides/sales.md - Locator: Sales agent examples and build links - Evidence kind: documented - Source date: Not stated - Retrieved: 2026-10-21