# MITHUNAI Docs

> Connect your documentation, repositories and files, and give your customers and teams grounded answers with citations.

- [AI answers from your own documentation](https://docs.mithunai.com/index.md): MITHUNAI turns the documentation, repositories and files you already have into an assistant that answers questions about them, with citations you can check.
- [Quickstart: your first cited answer](https://docs.mithunai.com/quickstart.md): Go from an empty organisation to a grounded answer: connect a source, wait for ingestion, create an assistant and ask it a question, using only the HTTP API.
- [Complete Enterprise Customer Onboarding Guide](https://docs.mithunai.com/get-started/onboarding.md): A complete step-by-step walkthrough for onboarding an enterprise organization to MITHUNAI: provisioning members, connecting knowledge, configuring assistants, evaluating grounding, and deploying production channels.
- [MITHUNAI Console Interface Walkthrough](https://docs.mithunai.com/get-started/console-walkthrough.md): A comprehensive visual guide to the MITHUNAI Web Console: Ask Playground, Knowledge Manager, Assistant Studio, Deployment Channels, Analytics, and Organization Settings.
- [Complete Enterprise Features & Capabilities Catalog](https://docs.mithunai.com/concepts/features-catalog.md): A comprehensive, searchable index of all capabilities, features, APIs, and architectural components delivered across the MITHUNAI enterprise platform.
- [How grounded answers and citations work](https://docs.mithunai.com/concepts/how-answers-work.md): Retrieval, grounding, citations and abstention: what happens between a question and an answer, and why the assistant says it does not know rather than guessing.
- [Organisations, roles and data isolation](https://docs.mithunai.com/concepts/organizations-and-roles.md): How MITHUNAI keeps each organisation's knowledge, assistants and conversations apart, and what an Owner, Administrator, Editor and Member may each do.
- [MithunAI Inference Gateway](https://docs.mithunai.com/concepts/gateway.md): High-throughput reverse proxy with sub-45ms semantic caching, automated prompt fencing, and resilient multi-provider model routing.
- [How your content becomes answerable](https://docs.mithunai.com/knowledge/overview.md): Collections, sources, ingestion jobs and documents: the four pieces that turn a website, a repository or an uploaded file into passages an assistant can cite.
- [Supported Knowledge Connectors & Data Sources](https://docs.mithunai.com/knowledge/connectors.md): Connect code repositories, websites, cloud storage, documentation spaces, and support platforms to your MITHUNAI knowledge base with automated delta synchronization.
- [Connect a website, sitemap or GitHub](https://docs.mithunai.com/knowledge/sources.md): Point MITHUNAI at a documentation website, an XML sitemap or a GitHub repository, and control which pages and paths it reads with include and exclude patterns.
- [Upload PDFs, Word documents and text](https://docs.mithunai.com/knowledge/file-uploads.md): Ingest content that lives on no website and in no repository: PDF manuals, Word documents, Markdown, plain text and source files, within the upload limits.
- [Sync sources and keep answers current](https://docs.mithunai.com/knowledge/ingestion.md): Ingestion reads a source, extracts its text, splits it into passages and indexes them. Start a sync, follow the job, read its counters, cancel a bad run.
- [Write content an AI can retrieve and cite](https://docs.mithunai.com/knowledge/content-best-practices.md): Answer quality depends more on your content than on the model. Structure pages under headings, state each answer plainly, and keep it current for retrieval.
- [Create and configure an AI assistant](https://docs.mithunai.com/assistants/configure.md): Choose the knowledge an assistant answers from, the model it runs on and the guidance that shapes its replies, then activate, disable or archive it later.
- [Supported Models, LLM Providers & Inference Engines](https://docs.mithunai.com/assistants/models-and-providers.md): Configure frontier and open-weights reasoning models across Anthropic, OpenAI, Google Gemini, DeepSeek, Groq, or private on-premises vLLM servers.
- [Assistant System Instructions & Prompt Guidance](https://docs.mithunai.com/assistants/prompt-guidance.md): How to write effective system instructions for MITHUNAI assistants: tuning persona, defining tone, constraining scope, and enforcing formatting standards.
- [Test an AI assistant before you launch](https://docs.mithunai.com/assistants/evaluate.md): Confirm three behaviours before real users arrive: it answers what your content covers, it abstains on what it does not, and it cites the right passage.
- [Automated Grounding Benchmarks & Hallucination Auditing](https://docs.mithunai.com/assistants/grounding-benchmarks.md): How MITHUNAI measures factual grounding, evaluates answer faithfulness, detects hallucinations, and executes automated golden test suites before production deployment.
- [MithunAI Agent Studio](https://docs.mithunai.com/concepts/agent-studio.md): Build stateful multi-agent workflows with tool authorization, directed graph execution, and human-in-the-loop approval gates.
- [Autonomous AI Agents, Tool Sandboxing & Execution Workflows](https://docs.mithunai.com/concepts/autonomous-agents.md): Deploy goal-directed multi-step agents with tool authorization, sandboxed container execution, durable state checkpointing, and human-in-the-loop approval gates.
- [Three ways to reach your assistant](https://docs.mithunai.com/channels/overview.md): One assistant, one body of knowledge and one grounding rule, reached three ways: an embeddable website widget, the HTTP API, and an MCP server for AI clients.
- [Add an AI chat widget to your website](https://docs.mithunai.com/channels/widget.md): Embed a chat assistant on your documentation or support site with one script tag. Anonymous visitors get grounded, cited answers, from the origins you allow.
- [Website Widget Theming, Customization & Embed Options](https://docs.mithunai.com/channels/widget-customization.md): Customize the MITHUNAI embeddable chat widget with your brand colors, custom typography, launcher positions, greeting prompts, and React components.
- [Build grounded answers into your product](https://docs.mithunai.com/channels/api.md): Call MITHUNAI from your own backend to ask questions, stream cited answers, manage knowledge and assistants, and read usage, with a server-held API key.
- [Official SDKs, Client Libraries & Code Examples](https://docs.mithunai.com/channels/sdks.md): Integrate MITHUNAI into Python, TypeScript, Node.js, and browser applications using our typed client libraries and streaming helpers.
- [MITHUNAI Command Line Interface (CLI)](https://docs.mithunai.com/channels/cli.md): Manage knowledge sources, trigger ingestion pipelines, query assistants, and run automated evaluation benchmarks directly from your terminal or CI/CD workflows.
- [Connect Claude and your IDE over MCP](https://docs.mithunai.com/channels/mcp.md): Let Claude Code, Cursor, VS Code and other Model Context Protocol clients search your knowledge and ask your assistants. Preview: three read-only tools.
- [Webhooks, Event Subscriptions & Notifications](https://docs.mithunai.com/channels/webhooks.md): Receive real-time HTTP callbacks when ingestion syncs finish, unanswered questions are flagged, or customer widget threads are escalated.
- [System Architecture & ADRs](https://docs.mithunai.com/concepts/architecture.md): Six-layer unidirectional dependency architecture and Architectural Decision Records governing safety and multitenancy.
- [Hybrid Retrieval, Reranking & Grounded Verification](https://docs.mithunai.com/concepts/retrieval-pipeline.md): MITHUNAI combines dense vector embeddings, BM25 sparse keyword search, cross-encoder reranking, and cryptographic prompt fencing to deliver sub-45ms grounded answers with verified citations.
- [Enterprise Use Cases, Architecture & Workflows](https://docs.mithunai.com/concepts/use-cases.md): How enterprise organizations implement MITHUNAI for developer documentation, internal engineering runbooks, customer support deflection, compliance auditing, and onboarding.
- [Team Members, Invitations & Access Management](https://docs.mithunai.com/administration/members.md): Manage organization membership, invite team members, assign role-based access control (RBAC) tiers, and audit administrative actions.
- [Create, rotate and revoke API keys](https://docs.mithunai.com/administration/api-keys.md): An API key lets a backend service, a script, a CI job or an MCP client call MITHUNAI with a role of its own. Create one, send it, rotate it, revoke it.
- [White-label the console and the widget](https://docs.mithunai.com/administration/branding.md): Present MITHUNAI under your own name, logo, colours and links, on the console your team uses and the website widget your visitors see, with no rebuild.
- [Usage reports and unanswered questions](https://docs.mithunai.com/administration/analytics.md): Measure how much your assistants are used and how often they answer, then read the questions they declined: each one names a gap your content could close.
- [Control what MITHUNAI support can see](https://docs.mithunai.com/administration/support-access.md): Support cannot reach your organisation until you allow it. What a session can read, how long it lasts, and the record it leaves in your own audit history.
- [Sovereign Cloud & Air-Gapped Deployment](https://docs.mithunai.com/administration/deployment.md): Deploy the complete MITHUNAI stack on private infrastructure, sovereign VPCs, or air-gapped environments with full data sovereignty and zero telemetry leaks.
- [Security, isolation and data handling](https://docs.mithunai.com/security.md): How MITHUNAI isolates each organisation's data, authorises every request, treats every ingested page as untrusted input, and what it does with your content.
- [Enterprise Compliance, SOC 2, HIPAA & Data Privacy](https://docs.mithunai.com/resources/compliance.md): MITHUNAI security controls, audit certifications, HIPAA BAA support, GDPR compliance, tenant data residency, and zero-training guarantees.
- [What is available, preview and planned](https://docs.mithunai.com/resources/product-status.md): Feature by feature: what is generally available, what is in preview and works but may change, and what is not built yet, so you can plan against the truth.
- [Troubleshooting: symptoms and fixes](https://docs.mithunai.com/resources/troubleshooting.md): Why the assistant abstains on everything, why the model list is empty, why an ingestion job fails, and what a 401, 403 or 429 means, each with the fix to apply.
- [Frequently asked questions about MITHUNAI](https://docs.mithunai.com/resources/faq.md): Does the assistant make things up? Is my content used to train models? Can another customer's data reach my answers? What evaluators and developers ask most.
- [Changelog: new features and changes](https://docs.mithunai.com/resources/changelog.md): Every MITHUNAI release, newest first: new knowledge sources, answer-engine changes, new channels, administration features, and changes to existing behaviour.
- [MITHUNAI HTTP API: base URL and rules](https://docs.mithunai.com/api-reference/introduction.md): The base URL, the versioning policy, the request and response conventions and the content types of the MITHUNAI HTTP API, and what every endpoint has in common.
- [Authenticate API requests with a key](https://docs.mithunai.com/api-reference/authentication.md): Send your API key as a bearer token on every request. Which credentials exist, what each may do, and which routes refuse a key and need a console session.
- [API error codes and how to handle them](https://docs.mithunai.com/api-reference/errors.md): Every MITHUNAI error carries a code and a message. Each code, the HTTP status it arrives with, what causes it, and what your client should do in response to it.
- [Rate limits and handling 429 responses](https://docs.mithunai.com/api-reference/rate-limits.md): The request budget for an API key, a signed-in person and a widget visitor, how each is counted, and how your client should read Retry-After after a 429.
- [Paginate list endpoints with cursors](https://docs.mithunai.com/api-reference/pagination.md): List endpoints use one of three paging shapes: cursor pages with a next_cursor, offset pages with has_more, and complete lists. Which endpoint uses which.
- [Stream answers with server-sent events](https://docs.mithunai.com/api-reference/streaming.md): Receive an answer as it is written, over a text/event-stream response. The event types and their order, how citations arrive, and how to end a stream early.
- [Start a conversation with an assistant](https://docs.mithunai.com/api-reference/conversations/create-conversation.md): Create an empty conversation bound to one assistant for life. The caller that creates it owns it, and only the owner may ask questions in it or rename it.
- [List the conversations you can see](https://docs.mithunai.com/api-reference/conversations/list-conversations.md): List conversations newest first. What you see is decided by your credential, not the query: an owner or admin sees all of them, every other role sees its own.
- [Get one conversation and its state](https://docs.mithunai.com/api-reference/conversations/get-conversation.md): Retrieve a conversation you own, or any of them if your role can moderate. One that is missing, in another organisation or hidden from you returns the same 404.
- [Rename, archive or restore a conversation](https://docs.mithunai.com/api-reference/conversations/update-conversation.md): Change a conversation's title, labels or archived state. Only the owner may: a moderator who can read someone else's conversation gets 404 when changing it.
- [Delete a conversation and its messages](https://docs.mithunai.com/api-reference/conversations/delete-conversation.md): Remove a conversation and every message in it. Deletion is irreversible, and only owner and admin hold it. To move one aside without losing it, archive instead.
- [Ask a question and get a cited answer](https://docs.mithunai.com/api-reference/conversations/send-message.md): Ask a question and receive the whole grounded answer, with its citations, in one response. The non-streaming turn, for work that cannot read an event stream.
- [Stream an answer as it is being written](https://docs.mithunai.com/api-reference/conversations/stream-message.md): Ask a question and receive the answer as Server-Sent Events as it is generated. The same turn as Send a message, with the same body and rules; delivery differs.
- [List the messages in a conversation](https://docs.mithunai.com/api-reference/conversations/list-messages.md): Read the questions and answers in a conversation, oldest first, paged by cursor. An answer is recorded before it is generated, so render only complete ones.
- [Create an assistant over your knowledge](https://docs.mithunai.com/api-reference/assistants/create-assistant.md): Create an assistant in your own organisation, pointed at the collections it answers from. The server assigns the id, and a new assistant starts out active.
- [List the assistants in your organisation](https://docs.mithunai.com/api-reference/assistants/list-assistants.md): List assistants as summaries, most recently updated first. A summary carries no instruction and a knowledge-base count rather than ids; archived ones omitted.
- [Get an assistant and its instruction](https://docs.mithunai.com/api-reference/assistants/get-assistant.md): Retrieve one assistant's full configuration, its instruction included. Archived ones are still returned; there is no delete route, so retire one by status.
- [Change an assistant's knowledge or model](https://docs.mithunai.com/api-reference/assistants/update-assistant.md): Change an assistant's name, knowledge collections, model, instruction or status. Send the revision you last read, so a concurrent edit is reported, not lost.
- [Get an assistant with its field limits](https://docs.mithunai.com/api-reference/assistants/get-assistant-settings.md): The same assistant object plus the field limits the server enforces and the allowed status values, so an editing form validates before it saves, not after.
- [List the models an assistant can use](https://docs.mithunai.com/api-reference/models/list-models.md): List the models this deployment can actually reach, which is what an assistant may use. A deployment with no provider configured can neither answer nor ingest.
- [List the model providers configured](https://docs.mithunai.com/api-reference/models/list-providers.md): List the model providers this deployment can route answers to. Providers are not part of MITHUNAI; a deployment reaches only those it holds credentials for.
- [Get one model from the catalogue by id](https://docs.mithunai.com/api-reference/models/get-model.md): Check a single model before setting it on an assistant. Every way this fails returns the same 404: unknown provider, no credential for it, or no such model.
- [Create a knowledge collection to ingest into](https://docs.mithunai.com/api-reference/knowledge/create-collection.md): Create the corpus your sources ingest into and your assistants answer from. The embedding model and vector width are fixed at creation and can never be changed.
- [List all your knowledge collections](https://docs.mithunai.com/api-reference/knowledge/list-collections.md): Every collection in your organisation, archived ones included, newest first, in a single unpaged response. Any role in the organisation can list them.
- [Get one knowledge collection by id](https://docs.mithunai.com/api-reference/knowledge/get-collection.md): Retrieve one collection, including the embedding model it was created with. An id that is unknown, one in another organisation and a malformed id all give 404.
- [Rename or archive a knowledge collection](https://docs.mithunai.com/api-reference/knowledge/update-collection.md): Rename, re-describe, archive or restore a collection. Sending null for a field is refused rather than treated as clearing it; send an empty string to clear one.
- [Create a knowledge source to ingest](https://docs.mithunai.com/api-reference/knowledge/create-source.md): Connect a documentation website, an XML sitemap, a GitHub repository or uploaded files to a collection. Creating a source stores it; it reads nothing yet.
- [List the knowledge sources you have](https://docs.mithunai.com/api-reference/knowledge/list-sources.md): Every matching source in one unpaged response, newest first, optionally for one collection. A well-formed id that is not yours matches nothing, not an error.
- [Get a knowledge source and its config](https://docs.mithunai.com/api-reference/knowledge/get-source.md): Retrieve one source and the configuration it was created with. Any role in the organisation can read it; for its latest run, ask for the source's latest job.
- [Delete a source and all its documents](https://docs.mithunai.com/api-reference/knowledge/delete-source.md): Disconnect a source and remove every document it produced, their versions, the indexed content and its jobs. The content stops being cited; nothing is undone.
- [Sync a source into its collection now](https://docs.mithunai.com/api-reference/knowledge/sync-source.md): Queue an ingestion run. It returns 202 as soon as the job is recorded, before anything is ingested, and is idempotent while a run for that source is active.
- [Get a source's latest ingestion job](https://docs.mithunai.com/api-reference/knowledge/get-source-job.md): The active job while a run is in progress, the most recent one afterwards, without needing a job id. A source never synced returns a null job, not an error.
- [List recent ingestion jobs and runs](https://docs.mithunai.com/api-reference/knowledge/list-jobs.md): Recent ingestion runs in your organisation, newest first. There is no cursor: this shows recent runs, not full history, and limit is clamped, never refused.
- [Get an ingestion job and its counters](https://docs.mithunai.com/api-reference/knowledge/get-job.md): Poll one ingestion job's status, attempt and document counters until is_active turns false. The counters update while the job runs and are final once it ends.
- [Cancel a running ingestion job early](https://docs.mithunai.com/api-reference/knowledge/cancel-job.md): Ask an active job to stop. Cancellation is cooperative: a running job stops between documents, and what it already ingested stays rather than being rolled back.
- [List the documents in a collection](https://docs.mithunai.com/api-reference/knowledge/list-documents.md): A page of the documents in one collection, newest first. Metadata only: no endpoint returns a document's text, and collection_id is required, never optional.
- [Upload files for a knowledge collection](https://docs.mithunai.com/api-reference/knowledge/upload-files.md): The only knowledge endpoint taking multipart/form-data. Uploading stores the files and adds nothing yet; post the returned configuration as an upload source.
- [Read the file upload size ceilings](https://docs.mithunai.com/api-reference/knowledge/upload-limits.md): The three ceilings file upload applies, so your client can refuse an oversized file before sending it. Deployment configuration: the figures the refusal uses.
- [Publish an assistant as a website widget](https://docs.mithunai.com/api-reference/widget-deployments/create-widget-deployment.md): Publish one assistant as an embeddable website widget and receive its widget key. The key is public by design; what protects the embed is its allowed origins.
- [List all your website widget embeds](https://docs.mithunai.com/api-reference/widget-deployments/list-widget-deployments.md): Every widget deployment in your organisation, newest first, revoked ones included so you can see an embed existed and was withdrawn. One page, and no cursor.
- [Get a widget embed and its widget key](https://docs.mithunai.com/api-reference/widget-deployments/get-widget-deployment.md): Read one deployment including its widget key, so you can recover the embed snippet at any time. The key is public; the origin allowlist is what protects it.
- [Change a widget embed's allowed origins](https://docs.mithunai.com/api-reference/widget-deployments/update-widget-deployment.md): Rename a deployment or replace the website origins allowed to host it. The assistant is fixed: to serve a different one, create a new embed and revoke this one.
- [Revoke a website widget embed for good](https://docs.mithunai.com/api-reference/widget-deployments/revoke-widget-deployment.md): Withdraw a deployment so its widget key stops authenticating. Revocation is immediate and permanent; to restore the widget, create a new embed and repaste it.
- [Bootstrap the widget on a visitor's page](https://docs.mithunai.com/api-reference/widget/bootstrap.md): Called from the visitor's browser before the launcher renders: confirms the embed is live and returns the brand to draw it with. Takes a widget key, no API key.
- [Start a widget thread for a visitor](https://docs.mithunai.com/api-reference/widget/start-thread.md): Open a conversation thread for one anonymous visitor and receive the visitor token that unlocks it. Takes the public widget key and allowed Origin, no API key.
- [List one visitor's own widget threads](https://docs.mithunai.com/api-reference/widget/list-threads.md): The conversation threads belonging to the visitor token you present, and no others. One visitor cannot enumerate another's threads by changing an identifier.
- [Get one widget conversation thread](https://docs.mithunai.com/api-reference/widget/get-thread.md): Read one thread from the visitor's browser, using the widget key and the visitor token the thread was started with. Both are needed; a key alone reaches none.
- [Ask a question from the website widget](https://docs.mithunai.com/api-reference/widget/send-thread-message.md): Ask a question in a widget thread and wait for the complete, cited answer. Takes the widget key, the visitor token, and an Origin the embed and platform allow.
- [Stream a widget answer to a visitor](https://docs.mithunai.com/api-reference/widget/stream-thread-message.md): Ask a question in a widget thread and receive the answer as Server-Sent Events while it is written, so a visitor watches it appear rather than waiting for all.
- [List the messages in a widget thread](https://docs.mithunai.com/api-reference/widget/list-thread-messages.md): Read the questions and answers in one widget thread, oldest first, for instance to restore the panel after a reload. Needs the widget key and visitor token.
- [MCP endpoint: JSON-RPC methods and tools](https://docs.mithunai.com/api-reference/mcp/mcp.md): Connect an MCP client to your organisation's knowledge over JSON-RPC 2.0 and call three read-only tools: search knowledge, list knowledge, ask an assistant.
- [Create an API key and read its secret](https://docs.mithunai.com/api-reference/api-keys/create-api-key.md): Mint a key and receive its secret exactly once. Takes a signed-in console session, never an API key, so a leaked key can never mint itself a replacement.
- [List your organisation's own API keys](https://docs.mithunai.com/api-reference/api-keys/list-api-keys.md): Every key, newest first, revoked ones included so an access review sees a key existed. Metadata only: no secret is returned, and a console session is needed.
- [Read an API key without its secret](https://docs.mithunai.com/api-reference/api-keys/get-api-key.md): Read one key's name, prefix, role, status and last use. The secret is never returned, by any route, at any time. Takes a console session rather than an API key.
- [Revoke a leaked or unused API key now](https://docs.mithunai.com/api-reference/api-keys/revoke-api-key.md): Stop a key authenticating, immediately and permanently, and read back when it happened. Takes a console session, so a key cannot revoke a key. Repeats are safe.
- [List your organisation's members](https://docs.mithunai.com/api-reference/members/list-members.md): Every seated person in the organisation, oldest seat first. An active session or an API key with member.read is required.
- [Seat a member in your organisation](https://docs.mithunai.com/api-reference/members/add-member.md): Seats an existing deployment account in the caller organisation. Requires a console session with member.invite.
- [Get usage counts for a time window](https://docs.mithunai.com/api-reference/analytics/get-usage.md): Conversation, question and answer counts for your organisation over a trailing or explicit window. Counts and rates only: no question or answer text is given.
- [List the questions nobody could answer](https://docs.mithunai.com/api-reference/analytics/list-unanswered.md): The questions your assistants declined, newest first, so you can see which gaps your content should close. Question text only: never the answer or who asked.
- [Read who from support can see your data](https://docs.mithunai.com/api-reference/support-access/get-support-access.md): Whether MITHUNAI support may open a session, who is inside now, and who has been. A support session cannot read this: the record of the access belongs to you.
- [Allow or withdraw MITHUNAI support access](https://docs.mithunai.com/api-reference/support-access/set-support-access.md): Turn MITHUNAI support access on or off. It is off until you turn it on, and withdrawing ends every live session at once rather than letting them run to expiry.
- [End one live MITHUNAI support session](https://docs.mithunai.com/api-reference/support-access/end-support-session.md): Stop a single support session while leaving support access on. It takes effect on that session's next request; a session in another organisation answers 404.
- [Read what a support session looked at](https://docs.mithunai.com/api-reference/support-access/get-session-activity.md): Every request the session made, newest first, refusals included. Read from your own audit log rather than from the session, so it outlives the session itself.
- [Get your organisation's brand overrides](https://docs.mithunai.com/api-reference/branding/get-branding.md): The overrides your organisation has set, not the fully resolved brand, plus the revision needed to change them. Your interface merges them over the defaults.
- [Get branding with its value limits](https://docs.mithunai.com/api-reference/branding/get-branding-settings.md): Everything the branding read returns, plus the value limits the server enforces and whether this caller may save. That flag is a display hint, not the control.
- [Replace your whole branding layer at once](https://docs.mithunai.com/api-reference/branding/replace-branding.md): A whole-layer replace, not a merge: the layer you send becomes the complete set of overrides, and anything you leave out falls back to the platform default.
- [Reset branding to the platform default](https://docs.mithunai.com/api-reference/branding/reset-branding.md): Clear every override in one step, so every surface shows the platform default again. Takes the revision from your last read, so a concurrent change is reported.
- [Health check for a load balancer probe](https://docs.mithunai.com/api-reference/operations/health.md): A liveness probe taking no credential and not rate limited. It does not check the database or any other dependency, and carries no version or configuration.
- [Readiness check for an orchestrator](https://docs.mithunai.com/api-reference/operations/ready.md): Whether this instance can reach the dependencies it needs. The status code is the verdict, so a plain probe suffices, and each dependency reports one word only.