A team-oriented multi-agent collaborative workforce platform supporting self-hosting and Kanban-driven task orchestration, transforming terminal-based AI coding tools into an observable and manageable workforce.
Core Positioning#
AgentsMesh is a self-hosted, team-level AI coding collaboration platform that transforms terminal-based AI tools into an observable and manageable workforce via AgentPods and multi-agent orchestration. Its capabilities focus on Agent lifecycle management, multi-agent collaboration orchestration, and end-to-end development task flows — not on underlying LLM training or fine-tuning.
Agent Execution & Isolation#
- AgentPod (Remote AI Workstation): Provides Web terminal, Git worktree isolation, and real-time streaming output, supporting multiple concurrent Pods.
- PTY Sandbox: The self-hosted Runner executes AI Agents in isolated PTY sandboxes for secure and independent execution environments.
Collaboration & Orchestration#
- Multi-Agent Collaboration: Coordinates multiple Agents via Channels and Pod Bindings, with real-time visualization of collaboration topology.
- Agent Plugin Architecture: Latest refactoring introduces Agent Plugin architecture with Loopal integration for improved extensibility.
- AgentFile (SSOT + ACP): Single Source of Truth (SSOT) architecture combined with Agent Control Plane (ACP) pipeline, paired with a CodeMirror editor.
Engineering Management & Task Flow#
- Kanban Task Management: Built-in Kanban board supporting Ticket-Pod binding and progress tracking.
- Code Hosting Integration: Supports GitLab, GitHub, and Gitee with MR/PR task linkage.
Enterprise Security & Governance#
- Multi-tenant Isolation: Organization > Team > User hierarchy with row-level data isolation.
- Access Control: Built-in SSO, RBAC, and audit logs.
- BYOK Model: Users bring their own API keys with no platform-imposed usage limits.
- Air-gapped Deployment: Supports fully air-gapped environments for high-security compliance scenarios.
Observability#
Integrated OpenTelemetry covering Backend, Runner, Relay and other core components for metrics and distributed tracing, easily connecting to Prometheus, Grafana and other monitoring tools.
Architecture Design#
Adopts a control-plane/data-plane separated architecture:
| Component | Tech Stack | Core Responsibility |
|---|---|---|
| Backend | Go (Gin + GORM) | API server: auth, org/team management, Pod lifecycle, task management |
| Web | Next.js (App Router) + TS + Tailwind | Frontend console: dashboard, Web terminal, Kanban, topology visualization |
| Relay | Go | Terminal relay cluster: low-latency WebSocket pub/sub bridging Runner and browser |
| Runner | Go | Self-hosted daemon: gRPC+mTLS to Backend, WebSocket to Relay, executes Agents |
| Web-Admin | Next.js | Internal admin panel: user/org/Runner management and audit logs |
Key Communication Mechanisms:
- Orchestration commands: gRPC + mTLS (Runner ↔ Backend)
- Terminal I/O streams: WebSocket (Relay ↔ Browser)
- Web authentication: JWT
- Reverse proxy: Traefik
- Data storage: PostgreSQL + Redis
- Object storage: MinIO (S3-compatible)
Installation & Deployment#
Option 1: Hosted Service Sign up at agentsmesh.ai, connect a Git provider, and launch Agents within minutes.
Option 2: Self-hosted Runner
curl -fsSL https://agentsmesh.ai/install.sh | sh
agentsmesh-runner login
agentsmesh-runner run
Supports curl script, deb, rpm, and Windows installers. Can also be installed as a system service for persistent operation.
Option 3: Docker One-click Dev Environment
git clone https://github.com/AgentsMesh/AgentsMesh.git
cd AgentsMesh/deploy/dev
./dev.sh
After launching the full stack, access the Web console at http://localhost:3000.
Agent Ecosystem Compatibility#
Natively supports Claude Code, Codex CLI, Gemini CLI, Aider, OpenCode, and is compatible with any custom terminal Agent.
Pending Information#
- Discord and LinkedIn specific links need to be extracted from the official website
- BSL-1.1 commercial license pricing is not publicly available (converts to GPL-2.0-or-later after 2030-02-28)
- Founding team background information is not explicitly listed in the repository
- Detailed ACP (Agent Control Plane) design mechanisms require further documentation review