Skip to content

Repository files navigation

Donna

Donna

A chat workspace where AI agents are real colleagues.
Part Slack, part Claude Cowork. Built for any business.

GitHub stars License: MIT Issues Status


Donna lives where your team already talks, channels, threads, direct messages, and adds AI agents that know your business because they read from your actual work: meetings, mail, documents, tickets, chats. Multi-tenant, self-hostable, open source.

Pre-1.0. Foundations are shipped (multi-tenant chat, connector framework, Fathom + Gmail ingestion). The smart context layer ("Cortex") and the retrieval agent are next. We are building in the open.

What Donna is

Shape Like Why
Chat workspace Slack, Discord The surface your team already knows: public + private channels, DMs, threads, mentions.
Collaborative AI workspace Claude Cowork Agents read, write, and respond inside the conversation. Draft together, refine in the thread.
Context substrate (the bet under both) One normalised layer over your tools so agents answer like teammates, not search bars.

Who it is for

  • Tech teams that want their coding agents grounded in real business context (MCP server included).
  • Service businesses (consultancies, agencies, ops, sales, marketing) that want AI colleagues who know the company.
  • Solopreneurs building a company where most of the "team" is well-designed agents handing work to each other.

What is built today

  • Multi-tenant chat workspace, channels, threads, DMs, membership.
  • Connector framework (provider / client / adapter / webhook / OAuth).
  • Bronze ingestion layer (raw artifacts, idempotent).
  • Fathom (meetings) and Gmail (mail) connectors, end-to-end.
  • Cortex (Silver) layer, structured queryable entities with edges, scoped clustering, anti-hallucination linter. In progress.

See the full roadmap for what is next: retrieval agent, Drive, draft/generation agent, WhatsApp, Linear + GitHub, private DMs with agents, learning personalities.

Repository layout

Path What it is Stack
server/ Backend, multi-tenant chat + agent runtime + ingestion + OAuth + admin Django 5.2+, DRF, Celery, Postgres 16 + pgvector, Redis 7, uv-managed Python 3.13+
web/ Web chat UI Vite + TypeScript
desktop/ Desktop chat client Electron + TypeScript
docs/ Public website (GitHub Pages) Static HTML
assets/ Brand assets, diagrams, figures PNG, SVG
server/plans/ Authoritative architecture + build plan Markdown

Quick start (local stack)

git clone https://github.com/cube-digital/donna.git
cd donna/server

# Full stack: Postgres + Redis + web + worker + beat
docker compose up --build

# Seed OAuth provider rows from env
docker compose run --rm web bootstrap

# Django shell
docker compose run --rm web shell

# Tail Celery worker
docker compose logs -f worker

Web UI: cd web && pnpm install && pnpm dev (http://localhost:5173).

Full setup, environment variables, and troubleshooting in server/plans/06-deployment-and-self-hosting.md.

Self-hosting

Donna runs the same code on Cloud and on-premise. Bring your own:

  • OAuth keys (via env vars or Django admin)
  • Storage (filesystem / S3 / GCS / Azure, S3-compatible services like MinIO and R2 work)
  • LLM (LiteLLM-fronted, any OpenAI-compatible endpoint)

One Postgres, one Redis, one Django app, one Celery worker. That is the stack.

Documentation

Where What
Website Vision, motivation, roadmap, changelog
server/plans/ Architecture, data model, conventions, deployment
server/plans/cortex/ The Cortex substrate, vision, contracts, examples, status
CLAUDE.md Repo-level guide for AI coding agents

Contributing

Donna is built in the open and we welcome contributions, big or small. Start with CONTRIBUTING.md for setup, conventions, and how to propose changes.

Quick links:

Community standards

License

MIT © Cube Digital.

You can run Donna for any purpose, including commercially, modify it, fork it, and self-host it. The only thing we ask is that the copyright notice stays with the code.

About

The context layer that connects your tools, your team, and your agents.

Resources

Contributing

Security policy

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Used by

Contributors

Languages