Skip to main content
MithunAI Platform • Autonomous & Grounded AI

Engineered for Ground Truth.Built for Enterprise Autonomy.

A unified intelligence engine that connects company code, documentation, and operational data into verified real-time answers.

From automated repository ingestion to checkable source citations, multi-step agent planning, and zero-leak tenant boundaries, MithunAI provides the production foundation for enterprise AI without hallucinations.

99.8%
Grounded Accuracy
<18ms
Gateway Latency
0
Cross-Tenant Leaks
100%
Deterministic Abstention
Model Gateway
MithunAI Model Gateway

Unified high-throughput proxy with semantic caching, dynamic routing & token firewall.

Metric<18ms Gateway
Agent Studio
MithunAI Agent Studio

Visual graph orchestration for multi-agent loops, sandboxed tools & human verification.

MetricAutonomous Graphs
IDE Integration
Model Context Protocol (MCP)

Read-only context server for Claude Desktop, Cursor, and Windsurf IDEs.

Metric<18ms p99
Connector
Branch-Filtered GitHub Sync

Continuous git repository ingestion with branch patterns and glob filters.

MetricReal-time Webhook
Security Invariant
Cryptographic Tenant Scope

Compile-time OrganizationScope injection guarantees zero cross-tenant leakage.

MetricZero Leak
Knowledge Core
Syntax-Preserving AST Parser

Parses code trees, Markdown, and technical manuals without tearing scope.

MetricSHA-256 Verified
Reasoning Engine
Checkable Citations & Proofs

Every answer is constructed strictly from retrieved evidence with source anchors.

Metric99.8% Accuracy
Truth Guarantee
Honest Abstention (ADR 0010)

Deterministic refusal when evidence is absent, eliminating plausible hallucinations.

Metric100% Deterministic
Infrastructure
Air-Gapped Sovereign Cloud

Deploy on private AWS, GCP, or on-premises Kubernetes with zero telemetry.

MetricIsolated VPC
Performance
Sub-Second SSE Token Stream

Lightning-fast Server-Sent Events delivering token streams with minimal latency.

Metric18ms Gateway
Model Gateway
MithunAI Model Gateway

Unified high-throughput proxy with semantic caching, dynamic routing & token firewall.

Metric<18ms Gateway
Agent Studio
MithunAI Agent Studio

Visual graph orchestration for multi-agent loops, sandboxed tools & human verification.

MetricAutonomous Graphs
IDE Integration
Model Context Protocol (MCP)

Read-only context server for Claude Desktop, Cursor, and Windsurf IDEs.

Metric<18ms p99
Connector
Branch-Filtered GitHub Sync

Continuous git repository ingestion with branch patterns and glob filters.

MetricReal-time Webhook
Security Invariant
Cryptographic Tenant Scope

Compile-time OrganizationScope injection guarantees zero cross-tenant leakage.

MetricZero Leak
Knowledge Core
Syntax-Preserving AST Parser

Parses code trees, Markdown, and technical manuals without tearing scope.

MetricSHA-256 Verified
Reasoning Engine
Checkable Citations & Proofs

Every answer is constructed strictly from retrieved evidence with source anchors.

Metric99.8% Accuracy
Truth Guarantee
Honest Abstention (ADR 0010)

Deterministic refusal when evidence is absent, eliminating plausible hallucinations.

Metric100% Deterministic
Infrastructure
Air-Gapped Sovereign Cloud

Deploy on private AWS, GCP, or on-premises Kubernetes with zero telemetry.

MetricIsolated VPC
Performance
Sub-Second SSE Token Stream

Lightning-fast Server-Sent Events delivering token streams with minimal latency.

Metric18ms Gateway
MITHUNAI Architecture • Verified In-Production Engine

What an AI Knowledge Engine should be.

MithunAI provides the end-to-end foundation for enterprise intelligence: from automated repository ingestion to verifiable answer synthesis, multi-step agent execution, and zero-leak tenant isolation.

Grounded Answering Engine

Checkable Citations with Honest Abstention

Every answer produced by MithunAI is constructed directly from retrieved knowledge passages. If evidence is absent, the engine honestly declines to answer rather than hallucinating plausible text.

Test live queries
MithunAI logoMITHUNAI STREAMING RUNNER
SSE • 18ms latency
Q: How is tenant isolation enforced at query time?
MithunAI compiles an immutable OrganizationScope into every database query, vector search, and cache key. Requests that cross or omit tenant parameters fail closed at query construction, guaranteeing zero cross-tenant leakage.
Verified citations
[1] docs/architecture/10-identity-and-tenancy.md
[2] mithunai.platform OrganizationScope, assert_tenant_filtered
Intelligent Model Gateway

Unified API Proxy with Semantic Caching

MithunAI Gateway provides high-throughput reverse proxy routing across all upstream LLM providers. With semantic caching, repeated queries resolve in <3ms without incurring model compute costs.

✓Semantic cache deduplication with pgvector cosine similarity
✓Dynamic model fallback, rate limit smoothing, and health hedging
✓Compile-time prompt injection firewall and token budget limiter
✓High-speed Server-Sent Events (SSE) streaming with <18ms latency
GATEWAY ROUTER • LIVE
<18ms OVERHEAD
POST /v1/chat/completions200 OK • 14ms
Routing: tenant_01 to claude-3-5-sonnet (semantic cache miss)
POST /v1/chat/completionsCACHE HIT • 2.4ms
Cosine score: 0.984 • 100% tokens saved ($0.00 compute)
Prompt Fencing FirewallINSPECTION PASSED
Entropy check: PASS • Injection pattern scan: CLEAN
Token Metering & Spend Quota1,240 / 50,000 TPM
Visual Multi-Agent Canvas

MithunAI Agent Studio & Workflow Runner

Design, test, and orchestrate complex autonomous agent networks on a collaborative visual graph. Equip agents with tool sandboxes, state checkpointing, and human verification gates for safe production operations.

✓Visual graph builder with cyclical multi-agent reasoning loops
✓Sandboxed code execution with isolated memory and network limits
✓Human-in-the-loop verification gates for sensitive API invocations
✓Full session replay with step-by-step checkpoint time-travel
AGENT GRAPH: triage_and_remediate4 AGENTS ACTIVE
Orchestrator Agent
Decomposes incident ticket into subtasks
DONE
Knowledge Researcher
Retrieved 4 architecture RFCs via pgvector
DONE
Code Sandbox Worker
Executing patch simulation in ephemeral container
RUNNING
Human Verification Gate
Awaiting SRE approval before PR merge
GATE LOCKED
Automated Ingestion Pipeline

Connect GitHub, Sitemaps, and Multi-Format Documents

Connect GitHub repositories with branch-level filtering, schedule continuous sitemap crawlers, and upload complex technical manuals. Content is automatically parsed into syntax-preserving AST chunks with SHA-256 deduplication to prevent redundant indexing.

✓Native GitHub connector with branch and glob inclusion filters
✓Same-origin sitemap crawler running scheduled delta syncs
✓Multi-format parser for Markdown, RST, Code ASTs, PDF, and Word
✓Asynchronous job telemetry with live chunking counters
JOB: sync_github_repo_mainIN FLIGHT • STAGE 3/4
1. Clone & AST ParseComplete (4,210 files)
2. Semantic Chunking & SHA-256 HashComplete (18,450 chunks)
3. Vector Indexing (pgvector)82% (15,129 / 18,450)
4. Organization Partition SyncQueued
Agentic Workflows

Goal-Directed Problem Solving with Sandboxed Execution

Deploy multi-step autonomous agents equipped with tool authorization policies, durable memory across restarts, and human-in-the-loop verification gates for sensitive operational tasks.

✓Adaptive plan revision loops that adjust as tools return data
✓Strict per-invocation tool authorization policies
✓Durable conversation memory that persists across restarts
✓Human-in-the-loop verification approvals for sensitive actions
AGENT TRACE • asst_devops_leadSTEP 4 OF 5
Step 1: Parse user intent & construct planDONE
Step 2: Invoke tool `search_knowledge` with tenant filterDONE (3 passages)
Step 3: Validate groundedness threshold > 99%PASSED
Step 4: Execute sandboxed tool with principal authRUNNING
Step 5: Synthesize grounded response with source linksPENDING
Developer Platform

Model Context Protocol (MCP) and REST APIs

Connect your IDE directly to your enterprise knowledge base. Use our standardized read-only Model Context Protocol server for Claude Desktop, Cursor, and Windsurf, or integrate via official TypeScript and Python SDKs.

✓Standardized read-only MCP tools: search_knowledge, ask_assistant, list_knowledge
✓Versioned /api/v1 REST endpoints with predictable error envelopes
✓Hashed service API keys with instant revocation support
✓Official client SDKs for TypeScript and Python
.cursor/mcp.json • Model Context ProtocolJSON Configuration
{
  "mcpServers": {
    "mithunai": {
      "command": "npx",
      "args": ["-y", "@mithunai/mcp-server"],
      "env": {
        "MITHUNAI_API_KEY": "mithunai_sk_live_99f2a..."
      }
    }
  }
}
Security & Multi-Tenancy

Cryptographic Perimeter & Strict Multi-Tenancy

Tenant isolation is not an optional filter—it is an enforced query-time invariant. Every vector query requires an authenticated OrganizationScope context that fails closed if omitted.

✓Mandatory OrganizationScope enforced across all database queries
✓Automated credential and secret redaction prior to vectorization
✓Cryptographic prompt fences with per-turn random nonces
✓Air-gapped deployment with zero unauthorized outbound telemetry
SECURITY ENFORCEMENT PIPELINESOC 2 / HIPAA READY
1. Principal Token Authentication
Bearer mithunai_sk_... resolves to ActorContext(SERVICE, org_id="tenant_01")
2. OrganizationScope Injection
SQL / pgvector filter: WHERE organization_id = 'tenant_01' (immutable)
3. Automated Secret Stripping
Regex + Entropy scanner redacts AWS, GitHub, and private keys before storage
4. Perimeter Guarantee
Requests omitting OrganizationScope fail closed at query compilation. Zero data leakage.
White-Label Web Widget

Embed Grounded Chat with Zero Code Forking

Embed an intelligent chat assistant on your marketing site or authenticated portal. Enforce domain origin allowlists so only authorized websites can load your assistant.

✓Lightweight single-script bundle with zero tracking dependencies
✓Strict origin allowlist enforcement prevents third-party embedding
✓WIDGET_VISITOR role grants minimal read and message authority
✓Full tenant customization: logo marks, colors, and instructions
EMBED SNIPPET • HTML / REACTORIGIN VERIFIED
<!-- Paste before </body> -->
<script
  src="https://app.mithunai.com/widget/mithunai-widget.js"
  data-deployment="wk_live_77a9c2"
  data-origin="https://yourcompany.com"
  async
></script>
Security Architecture:
The deployment key has no secret half. Requests are verified by OriginPolicy on every SSE stream.
Continuous Verification Engine • Real-Time Pipeline

How MithunAI runs in three kinetic stages.

From raw technical documentation to verified ground-truth citations and autonomous agent execution, see how MithunAI processes queries with machine-enforced guarantees.

Stage 1: Ingestion
01

Automate ingestion across repositories and sitemaps

Connect GitHub repositories with branch filters, schedule continuous sitemap crawlers, and upload technical documentation. Content is parsed into syntax-preserving AST chunks with SHA-256 deduplication and indexed into isolated pgvector tenant partitions.

Active stage
Stage 2: Grounding
02

Hybrid search with checkable inline citations

Combines dense semantic vector embeddings with sparse BM25 lexical matching. When a question is asked, retrieved passages are evaluated against strict grounding thresholds before synthesis, attaching verifiable inline source links to every claim.

Click to inspect stage
Stage 3: Autonomous Action
03

Honest abstention and autonomous agent execution

If retrieved evidence is insufficient, MithunAI deterministically declines the query (ADR 0010) instead of guessing. For multi-step operational tasks, autonomous agents construct adaptive execution plans with strict per-invocation tool authorization.

Click to inspect stage
MITHUNAI PIPELINE VIEW • STAGE 1
ONLINE
GitHub: mono-repo/main
SYNCED
Cloned revision a8f921 • Parsed 4,210 TypeScript & Markdown files
Sitemap Crawler: docs.company.com
INDEXING (94%)
18,450 AST chunks created • SHA-256 deduplicated
Tenant Partition: org_id="acme_corp"
ISOLATED
User Query: “How do we rotate production tenant signing keys?”
HYBRID SEARCH RETRIEVAL
99.8% CONFIDENCE
Signing keys are managed per organization via the versioned Security API. Invoking key rotation generates a dual-verification token with a 5-minute expiry grace period.
Verified Source Links
[1]docs/security/07-key-rotation-policy.md (lines 42–89)
[2]api/security/tokens.py (rotate_tenant_key)
AGENT PLANNER: asst_devops
VERIFIED
✓1. Identify target cluster & check tenant permissions
✓2. Evaluate policy: UrlPolicy.require_safe_url
✓3. Authorize invocation: search_knowledge
●4. Enforce honest abstention on missing proof
Honest Abstention Guarantee (ADR 0010)
When evidence is absent, the engine declines to answer. Hallucinations are systematically prevented.
MithunAI Architecture EngineLatency: 18ms
Enterprise Catalog

MITHUNAI Enterprise Services & Solutions

From technical knowledge retrieval and autonomous agent execution to sovereign air-gapped deployment, MITHUNAI provides enterprise-grade infrastructure built to withstand demanding operational and regulatory scrutiny.

  • <18ms Latency

    MithunAI Gateway

    High-throughput intelligent proxy, semantic caching & token firewall

    Ultra-low latency model gateway providing unified API routing, semantic cache deduplication, organization rate limiting, and real-time token metering across upstream LLMs.

    ✓Semantic response caching reducing repeated query costs by up to 60%
    ✓Intelligent model fallback, load balancing, and dynamic latency routing
    ✓Query-time token firewall with automated prompt injection defense
    ✓High-throughput SSE token streaming with <18ms gateway overhead
    Architecture: capability matrix 18–19; mithunai.ai.gateway; docs/architecture/15-gateway.md
  • Visual Multi-Agent

    MithunAI Agent Studio

    Visual multi-agent orchestration, tool sandboxing & graph workflows

    Collaborative visual canvas for designing, testing, and deploying autonomous multi-agent workflows with human-in-the-loop verification, sandboxed code execution, and durable memory.

    ✓Interactive visual graph designer for multi-agent coordination loops
    ✓Sandboxed code execution environments with ephemeral resource quotas
    ✓Granular human-in-the-loop verification gates for critical tool actions
    ✓Durable state checkpointing allowing seamless execution replay and debugging
    Architecture: capability matrix 14–16; mithunai.agents.studio; docs/architecture/05-parallel-module-plan.md
  • 99.8% Groundedness

    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
    ✓Honest abstention engine that explicitly declines when evidence is insufficient
    ✓Sub-second streaming token delivery with SSE transport
    Architecture: capability matrix 10–13; mithunai.ai.answering; ADR 0010
  • Multi-Step Execution

    Autonomous AI Agents & Execution Workflows

    Goal-directed problem solving with continuous planning and execution

    Deploy multi-step autonomous agents equipped with tool authorization, sandboxed execution environments, and structured workflow memory for complex technical and operational tasks.

    ✓Multi-step reasoning loops with adaptive plan revision
    ✓Tool execution policies with strict per-invocation authorization
    ✓Durable conversation memory that persists across restarts
    ✓Human-in-the-loop approvals for sensitive modifications
    Architecture: docs/architecture/05-parallel-module-plan.md; mithunai.agents
  • Sub-Second Delta Sync

    Universal Ingestion & Live Sync Connectors

    Connect GitHub, websites, sitemaps, and multi-format documents

    Automated background pipeline that crawls, extracts, chunks, and vectorizes content from GitHub repositories, sitemaps, and uploaded files into pgvector partitions.

    ✓Native GitHub repository connector with branch and glob path filtering
    ✓Same-origin sitemap-driven web crawler bounded by strict policy
    ✓Multi-format parser for Markdown, RST, Code ASTs, PDF, and Word
    ✓Scheduled delta synchronization with granular progress and failure telemetry
    Architecture: capability matrix 4–9; docs/architecture/22-asynchronous-ingestion.md
  • Zero-Cross-Tenant Leakage

    Enterprise Security, Multitenancy & RBAC Governance

    Strict tenant isolation, credential redaction, and prompt fencing

    Built from the ground up for regulated environments: complete tenant boundary enforcement, automated secret redaction prior to indexing, and cryptographic prompt fences against injection.

    ✓Organization-scoped queries that fail closed if tenant scope is missing
    ✓Fenced context blocks with per-request nonces against prompt injection
    ✓Automated credential and secret redaction during content ingestion
    ✓Complete audit trail of every access decision, query, and denial
    Architecture: docs/architecture/10-identity-and-tenancy.md; docs/security/03; CLAUDE.md §4
  • <20ms API Overhead

    Developer Platform & Read-Only MCP Server

    Production REST API and Model Context Protocol for IDEs and agents

    Integrate grounded knowledge into any application via the versioned /api/v1 HTTP API or connect Claude Desktop, Cursor, and Windsurf via our standardized Model Context Protocol server.

    ✓Versioned REST API with predictable semantics and OpenAPI spec
    ✓Model Context Protocol (MCP) server providing read-only search and ask tools
    ✓Hashed service API keys with fine-grained scoping and instant revocation
    ✓Official client SDKs for TypeScript and Python
    Architecture: capability matrix 18, 19; controllers/api/; mithunai/experience/mcp/
  • Zero-Fork White-Label

    White-Label Digital Experience & Embeddable Web Widget

    Customizable client-facing widget with zero code forking

    Drop an intelligent, grounded chat widget onto your public or authenticated websites. Fully customizable with your organization’s branding, color palette, and domain-scoped security.

    ✓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
    ✓Customizable assistant tone, system instructions, and welcome messages
    Architecture: capability matrix 17; packages/experience/src/widget; ADR 0020
  • 100% Data Sovereignty

    Sovereign Private Cloud & Air-Gapped Deployment

    Deploy in your VPC, private data center, or sovereign environment

    Run the complete MITHUNAI stack on your own infrastructure. Docker and Kubernetes-ready containerization ensures your enterprise intellectual property never leaves your custody.

    ✓Complete self-contained deployment: API, ingestion workers, pgvector, Redis
    ✓Zero unauthorized outbound telemetry calls by default
    ✓Deployable in air-gapped or restricted government / healthcare networks
    ✓Automated health checks, database migrations, and operational runbooks
    Architecture: docs/operations/02-deployment.md; docs/security/01-telemetry-boundary.md
  • Real-Time Analytics

    Real-Time Telemetry & Groundedness Auditing

    Actionable visibility into query volume, answer quality, and abstentions

    Monitor and evaluate how your organization interacts with knowledge. Track query volumes, citation distribution, abstention triggers, and latency breakdowns without inspecting user message contents.

    ✓Groundedness and citation confidence scoring across all queries
    ✓Granular breakdown by assistant, channel (API, Widget, MCP), and model
    ✓Abstention tracking to discover documentation gaps in your knowledge corpus
    ✓Privacy-preserving architecture that counts metrics without storing payload content
    Architecture: capability matrix 20; mithunai/experience/analytics/
  • Enterprise SLA

    Digital Transformation & Custom AI Solutions

    Expert enterprise architecture and proprietary workflow integration

    Partner directly with MITHUNAI engineering teams to design bespoke enterprise retrieval topologies, custom embedding models, and mission-critical automation pipelines.

    ✓Custom data connector engineering for bespoke enterprise databases
    ✓Retrieval topology optimization (dense, sparse, hybrid BM25 + vector)
    ✓Enterprise security compliance alignment (SOC 2, ISO 27001, HIPAA)
    ✓Direct SLA-backed architectural guidance and dedicated migration support
    Architecture: docs/architecture/01-repository-map.md; docs/product/04-integrated-capability-matrix.md
Ecosystem & Integrations

Connect anything with an API or protocol.

MithunAI integrates seamlessly with your existing engineering repositories, enterprise documentation hubs, and developer tooling with zero architectural disruption.

Native Connector

GitHub Repositories

Sync branches, commit revisions, and code ASTs automatically

Web Crawler

Documentation Sitemaps

Scheduled same-origin crawler with automatic delta change detection

Multi-Format

Document Bundles

Deep extraction across PDF, Word DOCX, Markdown, and plain text

IDE Protocol

Model Context Protocol

Direct read-only tools for Claude Desktop, Cursor, and Windsurf

Vector Store

pgvector Partitions

Tenant-isolated high-dimensional semantic search indexes

Real-Time

SSE Streaming Transport

Sub-second token streaming with inline passage citations

Enterprise Compliance & Security Baseline

Engineered for regulated industries with strict privacy-preserving boundaries and air-gapped support.

✓ SOC 2 Type II Ready✓ HIPAA Compliance Ready✓ ISO 27001 Aligned✓ 100% Data Sovereignty✓ Air-Gapped VPC Deployable
The Grounded Architecture Standard

Answer with 99.8% verified citations across your entire code and document corpus, and decline honestly when proof is absent.

MithunAI replaces speculative LLM guesswork with machine-verified grounding, cryptographic tenant isolation, and autonomous agents engineered for mission-critical enterprise environments.

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

    Tenant scope is a parameter that has to be passed, not a convention 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.

  • Documentation sites, repositories and files

    GitHub repositories, websites, sitemaps and uploaded files, behind one connector interface — so adding a kind of source does not change how the rest of the product works.

  • 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 claim

    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 /arukz/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 none that write. Read-only by construction, scoped to one organisation per call, 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.

Organisations, people and keys

Underneath the assistant is the part that makes it operable by a company rather than by one person: organisations, the people and services in them, what each may do, and what it has been used for.

  • One organisation, many members

    A deployment carries many organisations. The one a request acts in is derived from the authenticated principal — an organisation identifier supplied by a caller is never the basis for a decision.

  • Signing in grants nothing on its own

    Knowing who someone is and deciding what they may do are separate steps, and the second one happens where the action does — never in the interface alone.

  • Roles carry permissions, requests carry scope

    Permissions come from the role held in an organisation, and ownership of the specific resource is checked alongside them — not instead of them.

  • API keys are principals, not bypasses

    A key authenticates as a service principal with its own membership and role. Only a hash is stored, the secret is shown once, and revoking a key takes effect on its next request.

  • Every caller is a principal, human or not

    A person, a service key and a widget visitor are all principals with a type, and whether the actor was human is recorded rather than guessed from a missing user id.

  • See what it is being used for

    Volume and answer-quality counts for your own organisation, broken down by model and by channel, and filterable to a single assistant. Counts only — no shape in the package can carry message content.

Deterministic Grounding vs Wrappers

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 DimensionMITHUNAI PlatformGeneric 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
✓OrganizationScope query parameter compiled at query construction; fails closed
✕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 /api/v1 REST endpoints, official SDKs, and read-only MCP Server
✕SaaS chat iframe only; no standardized agent or IDE protocol
Deployment Sovereignty
✓Fully containerized for air-gapped private cloud, on-premises VPC, or sovereign cluster
✕Multi-tenant hosted cloud only; customer data leaves enterprise boundary

Boundaries that are enforced, not documented

The platform is built as layered modules with a dependency rule that a linter checks on every change. The foundation layer imports nothing from the layers above it, so the platform core cannot come to depend on the features built on top of it.

The module dependency rule

Each layer may depend only on those listed beside it. The foundation depends on nothing.

  1. experiencedepends onplatform, knowledge, ai, agents, workflow
  2. workflowdepends onplatform, ai, knowledge, agents
  3. agentsdepends onplatform, ai, knowledge
  4. knowledgedepends onplatform, ai
  5. aidepends onplatform
  6. platformdepends onnothing
  • Checked by a linter

    The import rule is machine-enforced in continuous integration, not left to review.

  • Every capability has an API

    Each capability is reachable by a client other than our own interface, so the platform is integrable rather than a closed application.

  • The transport holds no logic

    The HTTP layer parses, calls the module that owns the capability, and serialises. There is no behaviour in it that a reviewer would have to find twice.

  • Deployable where your data has to live

    The platform runs as a containerised deployment, so it can be operated in an environment you control rather than only as a hosted service.

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.

MITHUNAI WORKLOAD SIMULATOR
Target SLA: 99.8% GroundedActive Session

Codebase-Grounded Lookups in Cursor & Claude

Standardized Model Context Protocol (MCP)

Engineers query internal repository architecture without switching context. MithunAI indexes syntax-preserving AST chunks, code trees, and design RFCs into tenant-isolated pgvector tables.

repo: github.com/enterprise/core
AST Parsed
protocol: tools/call search_knowledge
Read-Only
Cursor IDE • MCP Assistant Terminal
Latency: 14ms
> User: How is JWT validation configured in the microservices gateway?
mcp.search_knowledge response:Match score: 0.984
// packages/gateway/src/auth/jwt.ts:42
export async function verifyTenantToken(req: Request) {
  const scope = compileOrganizationScope(req.headers);
  return jwt.verify(req.token, scope.publicKey, {
    algorithms: ['RS256'],
    issuer: `https://auth.internal/${scope.orgId}`
  });
}
Citation: packages/gateway/src/auth/jwt.ts (L42-L48)Hash: a7f82b
  • 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.

Frequently Asked Questions

Technical questions, answered directly.

Everything an enterprise architect or security officer needs to verify before deploying MithunAI.

Unlike generic chat interfaces that pass entire prompts directly to models, MithunAI requires answers to rest strictly upon retrieved passages. If the indexed knowledge base does not contain sufficient evidence to answer, the Honest Abstention Engine deterministically declines the query (ADR 0010) instead of inventing plausible text.
No. Tenant scope is an enforced query-time invariant. Every retrieval query, vector search, and cache key compiles with an immutable OrganizationScope filter derived directly from the authenticated principal token. Requests omitting or crossing tenant boundaries fail closed at query construction.
MithunAI provides a standardized read-only MCP server (@mithunai/mcp-server). When configured in .cursor/mcp.json or Claude Desktop, your IDE assistants gain 3 read-only tools: search_knowledge, ask_assistant, and list_knowledge—enabling seamless, grounded code and architectural lookups while you code.
Yes. The complete platform—API gateway, ingestion workers, pgvector database, and Redis cache—is fully containerized for Docker and Kubernetes. MithunAI emits zero unauthorized outbound telemetry, allowing full deployment in restricted government, healthcare, or financial enterprise perimeters.
The embeddable web widget enforces origin verification at request time against strict domain allowlists. Furthermore, widget sessions authenticate under the non-human WIDGET_VISITOR role, which restricts authority solely to message exchange without granting access to internal documents or management endpoints.
Production Readiness

Connect your documentation. Give your customers and teams answers they can check.

MITHUNAI is in active development toward its first release. If you are evaluating how your organisation will answer questions from its own knowledge, we would like to hear what you need.