desplega-ai/agent-swarm

agent‑swarm – an open‑source operating system for AI‑native work

What it is – agent-swarm is a framework that lets a “lead” AI agent receive tasks from everyday channels (Slack, email, GitHub, an API, etc.) and then delegate those tasks to specialised worker agents that run inside isolated Docker containers. The workers share a persistent “brain” (vector‑search memory and an identity store) so they can keep context across sessions, and the system can automatically create pull‑requests, Slack replies, or email responses on your behalf.

Key components

• Lead agent – plans, prioritises and routes incoming work.
• Worker agents – Docker‑isolated runtimes that can run any supported model (Claude Code, Codex, Pi, Open‑Code, Devin, ACP, etc.).
• Persistent brain – vector‑search memory + identity (SOUL, CLAUDE.md) that survives restarts.
• Integrations – ready‑made connectors for Slack, GitHub/GitLab, Linear, Jira, WhatsApp, Sentry, and more.
• Realtime rooms – shared state and presence channels for collaborative debugging or live monitoring.
• Extensible harness – you pick the underlying LLM provider; the platform is model‑agnostic.

Typical use‑cases

• Automating code reviews and PR creation directly from Slack or issue trackers.
• Running scheduled maintenance scripts (e.g., nightly dependency updates) with full context.
• Building internal “AI assistants” that can read tickets, write code, and reply to customers without human intervention.
• Providing a self‑hosted AI‑native backend for product teams that want to keep data on‑premise.

Quick start

# Add the skill to your existing operator (if you use the Desplega skill system)
npx skills add desplega-ai/agent-swarm

# Or run the repo directly
git clone https://github.com/desplega-ai/agent-swarm.git && cd agent-swarm
cp .env.docker.example .env   # fill in API_KEY, harness creds, and the eight agent UUIDs
openssl rand -base64 32 > encryption_key && chmod 600 encryption_key
docker compose -f docker-compose.example.yml --env-file .env up -d

The API will be available at http://localhost:3013 with Swagger docs at /docs. A web dashboard is also provided at https://app.agent-swarm.dev for visual monitoring.

Documentation & community

• Full docs: https://docs.agent-swarm.dev
• Live demo site: https://agent-swarm.dev
• Discord community for support and feature discussion.
• Contribution guide and open PR policy (see CONTRIBUTING.md).

License – MIT (2025‑2026, Desplega Labs). The project is actively maintained and welcomes community contributions.