MITHUNAI RAG · Live
AI answers from your documentation, with citations
MITHUNAI RAG turns your documentation, GitHub repositories, help centres and files into an assistant that answers with citations a reader can follow, or says plainly that your sources do not cover the question. Put it on your site as a widget, call it over the HTTP API, or reach it from Cursor and Claude Code through a read-only MCP server.
Last updated
See a real answer, and a real “I don’t know”
Captured from the public playground on 30 October 2026. The playground answers only from MITHUNAI’s own documentation.
“How do I add the Ask AI widget to my website?”
The widget puts an assistant on your own website, and visitors who have never signed in can ask questions and get grounded answers with citations from an Ask AI launcher button [1]. An Owner, Administrator or Editor creates one embed per site by opening Integrations → Website widget, selecting New embed, entering a Name, choosing the Assistant that should answer, listing your site under Allowed origins, and selecting Create embed; the embed then appears with its Embed snippet containing its widget key [2]. Widget requests must be allowed by two allowlists: your embed’s own list and a deployment-wide list set by whoever operates your MITHUNAI deployment; both start empty, so you must ask your MITHUNAI administrator to add your site’s origin to the deployment-wide list [2]. Add the snippet to your site by selecting Copy beside the embed’s Embed snippet in Integrations → Website widget and pasting it unchanged before the closing </body> tag on every page that should show the widget [2].
- [1] Add an AI chat widget to your website, #add-an-ai-chat-widget-to-your-website (88% match)
- [2] Add an AI chat widget to your website, #set-it-up (89% match)
Answered from the sources listed with it.
“What is the price of the Business plan in euros?”
I don’t know based on the available knowledge.
No answer from your sources. This assistant answers only from the documentation behind it, and that documentation does not cover this. Rewording the question may find it.
An assistant your customers can act on
The problem with putting a model in front of customers is not that it cannot answer. It is that it answers anyway when it should not. MITHUNAI is built around the opposite behaviour: an answer is checked against your own content before it is given, and withheld when that content does not support one.
Grounded in your content
Retrieval assembles evidence from your corpus, and the answer engine is required to build its response from that evidence rather than from what the model happens to recall.
Citations a reader can follow
Each answer carries the passages it was built from, so the person reading it can check the source instead of trusting the system.
An honest "I don’t know"
When the evidence does not support an answer, the engine abstains and says so. That is the product working. An assistant that answers everything is one you cannot trust on anything.
Isolated per organisation
Which organisation data belongs to is a value the code must pass, not a convention someone has to remember. Code that omits it does not compile into a working query. It fails closed.
Connect what you have already written
Point MITHUNAI at the content your organisation already maintains. It is fetched, extracted, split for retrieval and kept in step with the source, with the progress and the failures both visible rather than only the successes.
Sixteen kinds of source, one connector interface
GitHub repositories and their issues, pull requests and discussions; websites, sitemaps and uploaded files; your own question-and-answer pairs; OpenAPI schemas; public Zendesk help centres and Discourse forums; and, through a saved connection, Confluence Cloud, Notion, Jira Cloud issues, Jira Service Management requests and solved Zendesk tickets. Every kind goes through one connector interface, so adding one does not change how the rest of the product works.
Limit: Confluence, Notion, Jira and Zendesk tickets are tested against the vendors' documented responses and not yet tried against a real account; a full sync of GitHub discussions through a deployment has not yet been observed.
Private sources, with a connection saved once
Save a token once for your organisation and use it for private GitHub repositories, Confluence Cloud spaces, Notion pages, Jira projects and Zendesk tickets. A source that names a connection is read with that connection and nothing else: never anonymously, and never with a token that belongs to someone else.
Limit: Cloud only for Confluence. Built and tested, and not yet tried against a real account.
Edit, pause and resume a source
Change what a source reads, rename it, or pause it and resume it later, without deleting what it has already indexed.
The formats technical content actually comes in
Markdown, plain text, reStructuredText, source code, HTML, PDF and Word are extracted and chunked for retrieval, so a handbook and a header file both become answerable.
Crawling stays where you put it
Website ingestion is same-origin and sitemap-driven, so connecting one documentation site does not become a crawl of the internet.
Ingestion you can watch and re-run
Each sync reports what was discovered, what was indexed and what failed, and runs on a worker rather than blocking the request that started it.
Answers that show their work
A question goes to retrieval, retrieval assembles the evidence, and the answer is built from that evidence and checked against it. What arrives is an answer with its sources, or a clear statement that your content does not cover the question.
Grounded, and streamed as it is written
The response is required to rest on the retrieved passages, and it arrives token by token so a reader is not left waiting on a blank screen.
Citations on every answer
Each answer carries the passages it was built from, anchored back to the document and the source they came from.
Abstention is a distinct outcome
When the evidence does not support an answer, the engine says so, and the API reports that as its own result, never dressed up as an answer with no citations.
Assistants you configure
Name, description, model, corpus and system instruction, per assistant, read fresh on every turn. Operator guidance shapes tone and audience and cannot weaken the grounding rules.
Conversations that persist
History is stored durably and survives a restart, so a conversation is a record rather than a session that vanishes.
Put it where the question gets asked
The same assistant, the same corpus, the same grounding and the same citations, through whichever surface fits: a widget on your own site, a server-to-server API, or an MCP server the tools your engineers already use can connect to.
A widget for your own domain
An embeddable assistant that answers anonymous visitors on your site. A visitor is put through the same authorization pipeline as every other caller, and an embed answers only on the origins you list.
A versioned HTTP API
Every delivered capability is reachable over /mithunai/api/v1 by any client that speaks HTTP. Versioned from the first route, so the contract can outlive its first consumer.
A read-only MCP server
Connect an MCP-capable client to your knowledge through three tools that search, list and ask, and a fourth that rates an answer your key received. Nothing in your knowledge can be changed through it; every call is scoped to one organisation and authenticated with the same key as every other caller.
White-label without a fork
Each organisation sets its own name, logos, colours and typography, and the widget it serves picks them up. Branding is configuration a customer administrator edits, not a build you maintain a branch of.
Connect the sources and tools you already use.
MITHUNAI reads your repositories, documentation sites and files, and answers through the developer tools that speak MCP or HTTP.
GitHub Repositories
Ingest public repositories, or private ones with a saved connection (not yet tried against a real account)
Websites & Sitemaps
Same-origin crawl or sitemap, re-synced by hand or on an operator schedule
File Uploads
Extraction from PDF, Word DOCX, Markdown, HTML and plain text
Model Context Protocol
Read-only tools for Cursor, VS Code and Claude Code, in beta
pgvector Partitions
Tenant-isolated high-dimensional semantic search indexes
SSE Streaming Transport
Answers streamed as they are written, with inline passage citations
Security Baseline
What is in place today. MITHUNAI holds no third-party security certification yet.
Inside the console
The console was redesigned in October 2026 around one navigation rail in five groups, on a more compact type scale. This is what each destination on it does.
Dashboard
Dashboard
This month’s answers against your organisation’s allowance, as “X of Y answers used this month”, with a notice as you near it and another at the limit, beside the usage and answer-quality counts.
Configure
Sources
Connect any of the sixteen kinds of source, then edit, pause, resume and re-sync each one and watch what every sync indexed and what failed.
Limit: Confluence, Notion, Jira, Zendesk tickets and private GitHub repositories are not yet tried against a real account.
Customizations
Each assistant’s model, instructions and the knowledge it answers from, read on every turn rather than fixed when it was created.
Deploy
Playground
Ask what your customers would ask, and see the answer with its citations, or the reason the assistant declined.
Integrations
The website widget with its origin allowlist, the HTTP API and the read-only MCP server.
API keys
Keys that hold their own role. Revoking one takes effect on its next request.
Analyze
Conversations
Read back the conversations your assistants have had. Opening someone else’s is recorded in the audit trail.
Coverage gaps
The questions an assistant declined, grouped by meaning as well as wording, so you can see which missing page would answer the most of them.
Limit: The similarity threshold is not yet tuned on live questions.
Top questions
What people ask most, never who asked. A question is listed only once three different people have asked it, and can be downloaded as CSV.
Limit: Grouping by meaning is not yet tuned on live questions, and a period’s newest 2,000 questions are read.
Source analytics
Which sources and documents your answers cite, over the last 7, 30 or 90 days.
Workspace
Settings: Members
Invite colleagues by e-mail, change a member’s role, remove them, or hand ownership to another member. Removal takes their sessions, API keys and pending invitations with it.
Limit: Invitations are in beta: a real invitation has not yet been tried.
Settings: Connections
The tokens your private sources use, saved once for the organisation, rotated in place, and never shown again.
Limit: The sources that use them are not yet tried against a real account.
The Architectural Difference
Generic LLM integrations guess when they should decline, invent APIs when they should cite, and rely on reviewer vigilance for tenant isolation. MITHUNAI is engineered around machine-enforced guarantees.
| Capability Dimension | MITHUNAI Platform | Generic AI Chatbot Wrappers |
|---|---|---|
| Grounding Guarantee | ✓Query must be substantiated by retrieved corpus evidence before generation begins | ✕Generates answers from model parameter memory; prone to hallucinated facts |
| Source Citations | ✓Deterministic inline citations linking directly to exact passages in source docs or code | ✕None, or unverified URL references generated by the model |
| Abstention Behavior | ✓Distinct, intentional abstention event when evidence is absent or insufficient | ✕Always answers anyway, fabricating plausible technical claims |
| Tenant Isolation | ✓Every query carries its organisation, and is refused if it cannot | ✕Prompt-level instructions or client-provided filters vulnerable to bypass |
| Credential Redaction | ✓Automated secret, API key, and token stripping before text enters the vector store | ✕Raw indexing stores secrets in vector partitions, exposing them in answers |
| Developer Ecosystem | ✓Versioned /mithunai/api/v1 HTTP endpoints and a read-only MCP server, in beta | ✕SaaS chat iframe only; no standardized agent or IDE protocol |
| Deployment Sovereignty | ✓Containerized for your own cloud account or on premises, with no default telemetry | ✕Multi-tenant hosted cloud only; customer data leaves enterprise boundary |
Where a grounded assistant earns its place
These are the workloads the delivered platform supports today: grounded question answering over knowledge you connect, inside an organisation boundary.
Codebase-Grounded Lookups in Cursor & Claude
Engineers query internal repository architecture without switching context. MITHUNAI indexes source files, Markdown and design RFCs, chunked along their structure, inside your organisation.
// packages/gateway/src/auth/jwt.ts:42
export async function verifyTenantToken(req: Request) {
const scope = compileOrganizationScope(req.headers); // visitor-ok: a code sample
return jwt.verify(req.token, scope.publicKey, {
algorithms: ['RS256'],
issuer: `https://auth.internal/${scope.orgId}`
});
}Technical documentation assistance
Answer product and API questions from your own documentation, with citations a reader can follow back to the page, and an abstention when the docs genuinely do not say.
Engineering knowledge retrieval
Connect repositories and internal documentation so engineers can ask questions across them without the answer inventing an API that does not exist.
Support enablement
Give a support team a grounded assistant over the same documentation the product ships, where every answer shows its evidence.
Regulated and multi-tenant environments
Operate one platform across organisations where isolation, authorization and an audit record are conditions of running at all.
How a question becomes a cited answer
1Connect
Add a documentation site, sitemap, GitHub repository or file. It is fetched, split for retrieval and kept in step with the source.
2Ask
A question retrieves passages from your own sources only, inside your organisation and no other.
3Answer or decline
The answer rests on those passages and cites them. When nothing supports an answer, it declines and says what it searched.
Who it is for
Support, documentation and developer teams whose customers and colleagues ask questions the documentation already answers.
Read the documentation for this product or open it at app.mithunai.com.
Where it stands today
- It is also at its own address, rag.mithunai.com, in Beta.
- The MCP server is in beta: three read-only tools that search, list and ask, and a feedback tool that rates an answer your own key received. Nothing in your knowledge can be changed through it.
- Besides the chat panel, the widget can be a search box inside your page, a docked sidebar, or a deflector beside your support form that shows a cited answer before the form is sent and counts the requests it turned away. The three are in Beta: built and tested, not yet run on a real page.
- Answer quality is in Beta: score an assistant against an eval set of your own questions, for grounding, citations and the sources you expected. It has not yet run against a live model.
- Proposed answers is in Beta: for a group of questions your sources could not answer, it drafts an answer from those sources for a person to review and publish.
- Answering in Slack, in the thread with numbered sources or “I don’t know”, is in Beta and off until it is set up: it has been tested against a scripted Slack and is not yet tried against a real account. Answering in Microsoft Teams, in a personal chat or by mention, is in Beta likewise: off until it is set up, and it has not yet run against a real Teams tenant.
Questions people ask first
The other MITHUNAI products, on the same account
- MITHUNAI Chat
Live
A chat workspace for your team, with cited answers from your own knowledge.
- MITHUNAI Gateway
Live
One OpenAI-compatible endpoint in front of your model providers.
- MITHUNAI AI search visibility
Beta. Not yet run against a real AI search engine.
How AI search engines answer your buyers’ questions, and whom they cite.
- MITHUNAI Team
Beta. Access on request.
Employees for support, IT, DevOps, marketing, sales and finance that act only with approval.
Grounded answers from your own knowledge, a chat workspace for your team and one model gateway, on one account.
A 30-minute walkthrough on your own use case: connect a source, ask it questions, check the citations and see what it does when the answer is not there. Or ask the public playground now, with no account.