Open agent runtime

Turn coding agents into repeatable team workflows

Agent-Compose brings Codex, Claude Code, Gemini, and OpenCode under one open control plane for configuration, execution, scheduling, isolation, and credential boundaries

View demo Open GitHub ↗

Compose-style configIsolated runtimesEvents and schedulesRuntime LLM Facade

agent-compose / projectdaemon online

reviewerRUNNING

agents:
  reviewer:
    provider: codex
    driver: docker
    triggers:
      - event: pull_request

10:24:08 sandbox created
10:24:11 agent connected
10:24:42 review completed

Core capabilities

One control plane from config to execution

Bring coding agents, isolated runtimes, and automation triggers into declarative configuration your team can review

CONFIG

Declarative projects

Describe agents, workspaces, MCP servers, skills, volumes, and schedulers in agent-compose.yml

RUNTIME

Independent sandboxes

Choose Docker, BoxLite, or Microsandbox to define a clear execution boundary for each run

AUTOMATION

Schedules and events

Build repeatable automation with cron, interval, timeout, event, and webhook triggers

CREDENTIALS

Model credential boundary

Runtime LLM Facade keeps provider credentials on the daemon instead of passing them into guests

How it works

Declare once, keep it running

The daemon owns state and runtime lifecycle, the CLI and APIs expose a unified control surface, and agents execute inside isolated environments

Read the original essay ↗

01Declareagent-compose.yml

→

02Orchestratedaemon + CLI

→

03Runisolated sandbox

Use Cases

Building Agent-Compose with Agent-Compose

Explore showcase ↗

01 · FEATURED CASE

Let agents help build themselves

engineering-agent-workflows turns an Issue into a reviewable Draft PR while keeping each step inside explicit engineering boundaries

Issue→workflow agent→Draft PR

Blog

View article ↗

An agent is a program. The model is one of its instructions.

↗

Built in the open

Open source, visible evolution

Agent-Compose is licensed under AGPL-3.0 and currently in Public Preview APIs, runtime packaging, and deployment defaults may change, with GitHub as the source for releases and updates

View source ↗View releases ↗

Start composing

Build your first agent workflow

Explore the source and demo, or join the WeChat group to discuss deployment and usage

View demo GitHub ↗