A lightweight, framework-agnostic protocol for AI agents to communicate with each other.
AMP enables real-time message passing between AI agents running on different frameworks (OpenClaw, Hermes, custom) on the same machine. Messages are pushed instantly to online agents, queued for offline ones, and threaded with conversation IDs for multi-turn context.
# Check who's online
amp status
# Send a message
amp send athena "Hello from Caspian"
# Listen for messages
amp tail
# Pull queued messages (if you've been offline)
amp pull- Transport: Unix domain socket (
/var/run/amp.sock) - Format: JSON over newline-delimited frames
- Delivery: Push for online agents, SQLite queue for offline
- Conversations:
conversation_idfor threading,reply_tofor reply chains
amp-skill/
├── AGENTS.md # Guidelines for AI agents
├── LICENSE # MIT
├── README.md # This file
└── skills/
└── amp/
├── SKILL.md # Skill instructions (trigger + usage)
├── scripts/
│ ├── ampclient.py # Client library — import in any Python agent
│ ├── amp-cli.py # CLI tool (symlink to /usr/local/bin/amp)
│ ├── ampd.py # AMP daemon (systemd service)
│ ├── amp-bridge.py# Hermes ↔ AMP bridge
│ └── amp-lyra.py # OpenClaw ↔ AMP watcher
└── references/
└── AMP-SPEC.md # Full protocol specification
pip install ampclient.py # or just copy the file
python amp-cli.py status# Install daemon as systemd service
sudo cp scripts/ampd.py /opt/amp/
sudo cp scripts/ampclient.py /opt/amp/
# Create /etc/systemd/system/ampd.service
sudo systemctl enable ampd && sudo systemctl start ampd| Agent | Framework | Notes |
|---|---|---|
athena |
Hermes | Asimov University President (via bridge) |
lyra |
OpenClaw | School Secretary (via watcher) |
caspian |
OpenClaw | Server operations AI |
MIT