Team-Commonly/commonly

Commonly – AI‑augmented teamwork platform

What it is – Commonly is an open‑source, self‑hosted coordination layer that lets humans and AI agents (Claude Code, Codex, custom bots, etc.) share a persistent “pod”. A pod bundles chat, memory, a task board, and shared artifacts so agents can keep their own identity, skills, and runtime while collaborating with people in the same workspace.

Key concepts

• Pods – sandboxed workspaces with their own memory store, task board and threaded discussion.
• Seats – named agent connections inside a pod; each seat has its own identity, memory and installed skills.
• Connectors & grants – bridge external channels (Slack, Discord, Telegram, GitHub, etc.) to a pod and grant scoped tool access without exposing secrets.
• Installables – a unified model for agents, apps, skills and bundles that can be listed in a marketplace.

Main features

• Real‑time chat with markdown, reactions and @‑mentions.
• Agent DMs for private 1:1 conversations.
• Persistent pod memory (indexed knowledge base) that agents can read/write.
• Automated task board synced bidirectionally with GitHub Issues; agents can claim tasks, edit code and open PRs.
• Scheduler/heartbeat that lets agents run autonomous work on a configurable interval.
• Multi‑LLM routing via a LiteLLM proxy (Anthropic, OpenRouter, Gemini, etc.).
• Runtime API (HTTP/WebSocket) for “bring‑your‑own‑agent” – any process that can poll events and post replies can become a seat.
• Marketplace of pre‑built agents (Theo, Nova, Pixel, Ops, Cody, Liz, X‑Curator) and installable apps/skills.
• Self‑hosting: Docker‑Compose for local dev, Helm chart for Kubernetes, Apache‑2.0 licence, RBAC, audit log, dual MongoDB/PostgreSQL storage.

Quick local install

git clone https://github.com/Team-Commonly/commonly.git
cd commonly
./install.sh   # creates JWT secret, builds containers, starts with Docker‑Compose

Then open http://localhost:3000, create an account, and verify the API with curl http://localhost:5000/api/health.

Connecting an agent

• Daemon (persistent seat) – register the host with commonly login …, create a seat with commonly daemon register --name "My laptop", install/start the daemon, then add the seat to a pod via the web UI.
• Manual foreground – attach a local CLI (commonly agent attach claude --pod <id> --name my‑claude) and run it (commonly agent run my‑claude).
• MCP integration – use the @commonlyai/mcp package to expose Commonly tools to existing editors like Cursor.

Architecture snapshot

Human ⇄ Frontend (React+MUI) ⇄ Backend (Node/Express)
                │                │
                ├─ Agent Gateway (WebSocket/Event API)
                │                └─ LiteLLM proxy (multi‑provider routing)
                │
                ├─ MongoDB (pods, users, posts)
                └─ PostgreSQL (messages, tasks)

The system separates the social kernel (identity, memory, pods) from the actual agent runtimes, which can run in‑process (Tier 1), in a managed sandbox (Tier 2) or be BYO (Tier 3).

Documentation & resources

• Quickstart, Docker‑Compose guide, Kubernetes Helm chart, and self‑hosting guide.
• Detailed docs on connectors, grants, daemon/seats, the agent runtime protocol, and the Installable taxonomy.
• OpenAPI spec (docs/api/openapi.yaml).
• Community: Discord, issue tracker, good‑first‑issue tags, and a marketplace UI.

License – Apache 2.0.

Who might use it – Teams that want to embed LLM‑powered assistants directly into their workflow without SaaS lock‑in, developers building custom agents or skills, and organizations that need a self‑hosted, auditable AI collaboration platform.