An AI-powered enterprise assistant that answers questions across your company's tools (GitHub, Slack, Notion, HubSpot, and more) and can take actions on your behalf — all through a chat interface, with role-based permissions, citations, and a full audit trail.
- Conversational knowledge agent — ask natural-language questions ("what's the status of PROJ-123?", "summarize last week's Slack thread on the outage") and get answers grounded in your real company data, with source citations.
- Multi-agent orchestration — a LangGraph pipeline classifies intent, selects relevant sources, checks permissions, retrieves data, plans tool calls, executes them, and validates citations before responding.
- 30+ SaaS connectors — real implementations for GitHub, Slack, Notion, and HubSpot, with stub connectors ready to wire up for Jira, Salesforce, Zendesk, Google Workspace, Confluence, and more.
- Per-user "Connect" flows — each user logs in and connects their own accounts from the UI (no code or env-var edits required); credentials are encrypted at rest and scoped per user.
- Role-based access control — Admin, Engineer, Analyst, and Viewer roles gate which connectors and which actions (read/write/delete/send) each user can use.
- Human-in-the-loop actions — write actions (e.g. creating a GitHub issue, posting to Slack) can require explicit human approval before executing.
- Audit logging & observability — every query and action is logged; LangSmith and OpenTelemetry tracing are built in for debugging and monitoring agent behavior.
- Streaming responses — answers stream token-by-token with live visibility into which agent step is currently running.
- Sign up / log in at
/loginwith email and password (powered by Supabase Auth). - Connect your accounts at
/connectors— click "Connect" next to GitHub, Slack, Notion, HubSpot, etc. to authorize via OAuth (handled by Composio). Disconnect anytime. - Chat at
/— ask a question or request an action. The agent figures out which connectors are relevant, checks you have permission, retrieves and cites sources, and streams back an answer. - Approve actions — if the agent wants to take a write action (e.g. create an issue, send a message), it pauses for human approval before proceeding.
- Audit trail — every interaction is recorded for compliance and debugging.
Backend
- FastAPI (Python) — REST API and SSE streaming
- LangGraph + LangChain — multi-agent orchestration pipeline
- Anthropic Claude / OpenAI / Gemini / Groq — pluggable LLM providers
- Supabase (Postgres + pgvector) — auth, vector store for RAG, audit logs, per-user connector credentials
- Composio — hosted OAuth for per-user connector "Connect" flows
- LangSmith + OpenTelemetry — tracing and observability
- Redis — caching
Frontend
- Next.js 15 (React 19, TypeScript)
- Tailwind CSS
- Supabase JS client (auth session management)
- Vercel AI SDK primitives for streaming chat UI
Infra
- Docker Compose (backend, frontend, Redis, OpenTelemetry collector)
- One place to ask, instead of ten tabs to search. Stop hopping between GitHub, Slack, Notion, and your CRM to piece together an answer — ask once and get a cited, synthesized response.
- Self-service, not IT-gated. Anyone can log in and connect their own accounts through the UI — no waiting on an admin to hardcode API keys or redeploy the backend.
- Safe by default. Role-based permissions and human-approval gates mean the agent can be genuinely useful (including taking actions) without becoming a liability.
- Trustworthy answers. Citation validation means responses point back to the real source document, not a hallucinated summary.
- Built to extend. New connectors follow a simple base-class interface, and the agent pipeline is a swappable LangGraph — add a data source or a new reasoning step without rearchitecting the system.
- Observable from day one. Tracing and audit logs are wired in up front, so you can debug agent behavior and demonstrate compliance instead of bolting on observability later.