AgentMesh is a self-hostable control plane for AI-agent teams. Give it a goal and acceptance criteria; assign specialist employees; then follow their work, handoffs, decisions, and results in one Console. Simple work can stay with one agent.
Status: Alpha. The supported scope is a single-team evaluation or non-critical deployment, not a production-grade multi-tenant or high-availability service. See the implementation status and v1 boundaries.
Docker is the only prerequisite. The default deterministic executor needs no model API key.
git clone https://github.com/0YHR0/AgentMesh.git
cd AgentMesh
docker compose up --buildOpen the Console, create a Direct task, and inspect its result. To use a real model, employees, and a coordinated task, follow the guided tutorials below. Do not enter provider keys on an unauthenticated public HTTP deployment.
| I want to… | Read |
|---|---|
| Create my first task without a key | Five-minute guide · 中文 |
| Configure DeepSeek and watch three employees collaborate | Real-model walkthrough with screenshots · 中文 |
| Set up model connections and employee memory | Model and memory setup · 中文 |
| Deploy, configure gates, or operate the service | Administrator best practices · 中文 |
| Understand architecture or extend the platform | Documentation map · 中文 |
The documentation map also links scenarios, API and feature references, operations runbooks, architecture decisions, proposals, and the roadmap.
- Direct, reviewed, and coordinated task execution with versioned agents and durable results.
- A Console for task creation, status, intervention, and a replayable Mission Map.
- Optional MCP tools, A2A delegation, human approval, budgets, company records, and memory.
- PostgreSQL as the business source of truth; Redis Streams for delivery. LangGraph is an execution adapter, while feature gates keep advanced capabilities opt-in.
See Contributing, Changelog, and the Apache 2.0 license.