AgentWeave¶
A collaboration framework for N AI agents — Claude, Kimi, Gemini, Codex, Minimax, GLM, and more.
AgentWeave lets multiple AI agents work together on the same project through a shared protocol. The AgentWeave Hub is a self-hosted server with a web dashboard — the recommended way to run it.
Quick Start¶
Get up and running in 5 minutes:
# 1. Install the CLI
pip install "agentweave-ai[mcp]"
# 2. Start the Hub
agentweave hub start
# 3. Initialize your project
cd /path/to/your-project
agentweave init --project "My App"
# 4. Activate (connects Hub, sets up MCP, starts watchdog)
agentweave activate
See the Getting Started guide for full details.
What is AgentWeave?¶
AgentWeave solves a simple but important problem: how do you get multiple AI agents to collaborate on the same codebase?
It provides:
- A shared protocol — tasks, messages, and context files that all agents can read and write
- Multiple transport modes — local filesystem, git orphan branch, or HTTP via the Hub
- An MCP server — native tool integration so agents can send messages and manage tasks autonomously
- A web dashboard — real-time visibility into agent activity, tasks, and messages
Three Modes of Operation¶
| Mode | Setup | Best For |
|---|---|---|
| Hub | Docker + HTTP transport | Teams, multi-machine, web dashboard (recommended) |
| Zero-relay MCP | agentweave mcp setup + watchdog |
Autonomous loops, same machine, no server |
| Manual relay | Zero setup | Quick one-off delegation |
Documentation¶
| Section | Description | Links |
|---|---|---|
| Getting Started | Install the CLI, start the Hub, and connect your first agents | Installation · Quick Start |
| Guides | Step-by-step guides for common tasks and workflows | Adding New Agents · Context Files · Session Modes · AW-Spec Workflow · AI Jobs · Pilot Mode · Dashboard · FAQ |
| Reference | CLI commands, MCP tools, API endpoints, and configuration options | CLI Commands · MCP Tools · Task Lifecycle · Hub API |
| Architecture | Understand how AgentWeave works under the hood | Overview · Transport Layer · Messaging · Locking |
| Contributing | Development setup and release process | Development · Release Process |
Dashboard Features¶
Open http://localhost:8000 to see:
- Mission Control — centralized overview of session status and activity
- Tasks board — all tasks with status, priority, assignee, and deliverables
- Messages feed — inter-agent messages with inline task linking
- Human questions — answer agent questions directly in the UI
- Agent activity — live event stream, per-agent output logs, and session management
- Agent chat — per-session chat history with session selector
- Agent cards — connected agents with multi-role badges, yolo mode, and runner type
See Using the Dashboard for more.