- 01
Onboard with your dedicated architect
We review your documentation sites, repositories and files with you, and decide which sources each assistant should answer from.
- 02
Connect your sources
Connect any of sixteen kinds of source, from repositories and websites to API references, help centres and files. What is indexed stays inside your organisation and is never searched on anyone else’s behalf.
- 03
Put answers where people ask
Deploy assistants through the website widget, the HTTP API and the read-only MCP server, with answers streamed as they are written and cited to their sources.
Self-hosted
Run MITHUNAI in your own cloud or data centre
The Sovereign deployment runs the whole platform (API, ingestion workers, a pgvector database and Redis) in your own cloud account or data centre. It sends no telemetry by default and calls model providers only with credentials you configure. Each organisation's data stays inside that organisation.
Last updated
Reference Architecture
One organisation, its sources, its assistants, and the channels that reach its people.
Each organisation connects its own sources, builds assistants over them, and delivers answers through a widget, the API or MCP. Everything it stores and retrieves stays inside its own organisation.
Key facts
16
Knowledge Source Kinds
3
Places To Ask
Read-only
MCP Tools (beta)
Demo to production in 3 phases
From the first source you connect to assistants your whole company uses, our engineering team works with you at every step.
Multi-layered, enterprise-grade security
Built so a security reviewer can check each control against the code that enforces it.
✓
Multi-Tenant Partition Isolation
Every query and embedding lookup is bound to the organisation of the signed-in caller, and a request without one fails closed.
✓
Automated Credential Redaction
API keys, tokens, and credentials are identified and stripped prior to chunking and vector storage, preventing accidental disclosure in citations.
✓
Prompt Fencing & Injection Defense
Retrieved passages are fenced with a per-turn random nonce and marked as data, which makes injected instructions in your documents harder to act on.
✓
Private Cloud & On-Premises Custody
Run the containerized platform in your own cloud account or data centre, with no outbound telemetry by default.
Enterprise Catalog
MITHUNAI Enterprise Services & Solutions
From ingestion and grounded answers to delivery channels and self-hosted deployment, MITHUNAI provides the infrastructure to answer questions from your own knowledge.
- One /v1 endpoint
MITHUNAI AI Gateway
One OpenAI-compatible endpoint with an encrypted key vault and fallback
A model gateway that puts one OpenAI-compatible /v1 endpoint in front of the providers you register. Provider keys are held in an encrypted vault, each gateway key has its own usage and budget, and a failed call falls back to another configured model.
✓OpenAI-compatible /v1 endpoint for completions and embeddings✓Fallback to another configured model when a call fails✓Gateway keys with their own usage and budget✓Provider keys held in an encrypted vault, not in each surface - Beta
Task Agents
Agents that plan over several steps and call tools
A surface for agents that act rather than only answer. Beta. Access on request. It runs behind sign-in, and sign-up is closed by design.
✓Deployed on its own address, behind sign-in✓Authentication is on by default and cannot be switched off by configuration✓Access on request - Chat Workspace
AI Chat Workspace
A conversational workspace over the same account and the same gateway
A conversational workspace for teams: threads, prompts, bookmarks and memories. It signs in through the platform, holds no model key of its own, and reaches providers only through the gateway.
✓Threads, a prompt library, bookmarks and memories✓Signs in with your platform account, with no password of its own✓No provider key of its own: every Chat model call goes through the gateway✓File uploads restricted to an allowlist of types - One Sign-In
One Sign-In & Staff Platform Operations
One account for the surfaces that support it, with a separate staff entrance
One identity, issued by the platform’s own OpenID Connect provider with PKCE required of every client. Chat signs in through it. Agents and the Gateway offer it too, and each still also shows a local sign-in form of its own. Staff reach the console through a separate entrance that checks the same account.
✓One OpenID Connect identity, with PKCE required of every client✓A staff entrance to the console that uses the same account, not a second one✓Signing in once and signing out once across every product: in beta✓Roles per organisation, with ownership of the resource checked alongside✓Planned: SSO with your identity provider, SCIM and fine-grained RBAC - Cited or Abstains
Enterprise Knowledge & Answer Platform
Grounded AI over your internal documentation, codebases, and files
Transform technical documentation, GitHub repositories, and internal handbooks into an intelligent assistant that answers questions with verified citations and declines when evidence is absent.
✓Grounded response generation required to rest on retrieved passages✓Interactive inline citations linking directly to exact source passages✓An honest “I don’t know” when the evidence is not there✓Streaming answer delivery over Server-Sent Events - 16 Source Kinds
Knowledge Ingestion & Source Connectors
Sixteen kinds of source, public and private
Background pipeline that fetches, extracts, chunks, and embeds content from GitHub repositories, websites, sitemaps, and uploaded files into pgvector, inside your organisation.
✓GitHub repositories, issues, pull requests and discussions; websites, sitemaps and files✓OpenAPI schemas, custom question-and-answer pairs, public Zendesk help centres and Discourse forums✓Private GitHub, Confluence Cloud, Notion, Jira, Jira Service Management and Zendesk tickets through a saved connection (built; not yet tried against a real account)✓Same-origin website crawler and sitemap source, bounded by a safe-fetch policy✓File upload for PDF, DOCX, Markdown, HTML and TXT✓Manual or operator-scheduled re-sync with job progress and failure counts - Fails Closed
Enterprise Security & Multitenancy
Strict tenant isolation, credential redaction, and prompt fencing
Tenant boundary enforcement on every query, automated secret redaction before indexing, and nonce-fenced prompts that make injected instructions harder to act on.
✓Every query is bound to one organisation, and refused if that cannot be established✓Fenced context blocks with per-request nonces against prompt injection✓Automated credential and secret redaction during content ingestion✓Audit events for security-relevant actions such as authentication and key use - MCP Beta
Developer Platform & Read-Only MCP Server
Production REST API and Model Context Protocol for IDEs and agents
Integrate grounded knowledge into any application through the versioned /mithunai/api/v1 HTTP API, or connect Cursor, VS Code and Claude Code to the remote Model Context Protocol server, in beta.
✓Versioned HTTP API with predictable semantics and a reference page per route✓Model Context Protocol (MCP) server providing read-only search and ask tools✓Hashed mithunai_sk_ API keys, each with its own role, revocable at any time✓No SDK required: call the API with curl or any HTTP client - Zero-Fork White-Label
White-Label Digital Experience & Embeddable Web Widget
Customizable client-facing widget with zero code forking
Drop a grounded chat widget onto your public or authenticated websites, carrying your organization’s name, logos and colour palette, and answering only on the origins you list.
✓Lightweight embeddable script with zero external tracking dependencies✓Configurable tenant branding: logos, custom colors, and typography✓Strict origin-checking pipeline ensuring widget only serves allowed domains✓Assistant name, model and system instruction set per assistant - Self-Hosted
Private Cloud & On-Premises Deployment
Deploy in your own cloud account or data centre
Run the MITHUNAI stack on your own infrastructure. Containerized services keep your documents and index in your custody.
✓Self-contained deployment: API, ingestion workers, pgvector, Redis✓No outbound telemetry calls by default✓Model providers called only with credentials your deployment configures✓Health and readiness probes, database migrations, and operational runbooks - Usage Analytics
Usage & Answer Analytics
Actionable visibility into query volume, answer quality, and abstentions
See how your organization uses its knowledge: question volumes, which sources the answers cite, and what went unanswered, as counts that carry no message content.
✓Usage counts for questions, answers and abstentions✓Breakdown by model and by channel, filterable to one assistant✓Abstention tracking to discover documentation gaps in your knowledge corpus✓Counts only: the analytics carry no message content - Hands-On
Solution Engineering
Hands-on help with content, retrieval and deployment
Work directly with MITHUNAI engineers on how your content is structured, how retrieval is tuned for it, and how the platform is deployed.
✓Guidance on structuring your content for grounded answers✓Retrieval tuning across vector and lexical ranking✓Answers for your security review questionnaire✓Direct architectural guidance and migration support
Frequently Asked Questions
Running it in your own environment
Yes. The platform (API, ingestion workers, a pgvector database and Redis) is containerised, and it sends no telemetry by default. It still needs to reach a model provider, with credentials your deployment configures, to embed documents and write answers.
Production Readiness
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.