Compozy
A free, open-source "operating system" that lets different AI coding agents — Claude Code, Codex, Gemini CLI, Cursor — work together on the same task instead of each running in its own isolated silo.
🔗 Visit Compozy
Description
If you use several different AI coding agents, they typically can't coordinate with each other at all — each runs standalone, with no shared state or awareness of what the others are doing. Compozy sits underneath all of them as a coordination layer, giving them shared state, scheduled execution, and a common way to hand off or collaborate on tasks.
Compozy is an open-source (MIT) daemon that acts as a local-first "operating system" for AI agents, storing state in SQLite and running as a single binary. It supports durable sessions that survive restarts, automated execution triggered by cron schedules, webhooks or events, centralized state management with human approval gates, and a "Compozy Network" protocol for peer discovery between agent instances. It exposes a Web UI, CLI, HTTP/SSE, and an MCP interface, and works with Claude Code, Codex, Gemini CLI and Cursor. It's a young project (created March 2026) with 2,663 GitHub stars.
💬 Our review
The short version: Compozy is an interesting bet on a real gap — today's AI coding agents don't talk to each other — but it's early enough that you should treat it as something to experiment with, not depend on for critical workflows yet.
The core differentiator is that it doesn't try to be another AI agent itself; it's infrastructure that coordinates the agents you already use, so Claude Code, Codex, Gemini CLI and Cursor can share state and hand off tasks through a common daemon rather than you manually copy-pasting context between them. The local-first, single-binary, SQLite-backed architecture also means no external dependencies or cloud service to trust — everything runs on your machine. Scheduled and event-triggered execution (cron, webhooks) opens the door to genuinely automated multi-agent workflows rather than only interactive sessions.
The honest limits: at 2,663 stars and roughly five months old, it has a nascent ecosystem with limited production examples to learn from, and coordinating multiple AI agent CLIs is inherently complex — expect setup friction. The "Compozy Network" peer-discovery protocol is also an emerging, non-standard approach that could evolve significantly. For a technically adventurous developer or team already juggling multiple AI coding agents who wants to experiment with real coordination between them, it's worth a try; for most people, sticking with one agent at a time remains simpler today.
💰 Pricing
📊 Global score
🤖 AI-enriched data
Outil open-source (MIT), entièrement gratuit et auto-hébergé (binaire unique, SQLite local).
Pros
Coordonne plusieurs agents IA existants (Claude Code, Codex, Gemini CLI, Cursor) plutôt que d'en être un de plus
Architecture local-first, binaire unique, SQLite — aucune dépendance externe
Exécution automatisée déclenchée par cron/webhooks/événements
Gratuit, open-source (MIT), sessions durables qui survivent aux redémarrages
Cons
Écosystème naissant (2,7k étoiles, ~5 mois d'existence) avec peu d'exemples de production
Protocole "Compozy Network" encore émergent, non standardisé
Coordination multi-agents intrinsèquement complexe — friction de mise en place attendue