Agent Relay is a low-latency messaging and coordination layer for AI agents.
Think: agents that actually talk to each other. In real time. Without rewiring your setup.
Agent Relay is a developer-first protocol and runtime that lets AI agents communicate like a team instead of a bunch of isolated scripts.
If your agent can print to stdout, it can participate.
Right now, most agents:
- Run alone
- Canāt coordinate
- Only "communicate" after theyāre done
Thatās not how teams work. And agents are teams.
Agent Relay gives them a shared, fast, boring-in-the-best-way communication layer.
- ā” Sub-5ms message delivery
- š Works across providers (Claude, Codex, Gemini, custom)
- š§ No config changes to your existing agent setup
- š„ļø CLI-native ā pipes, streams, stdout, not frameworks
No YAML marathons. No SDK lock-in. No magic.
Youāll find:
- Core protocol implementations
- CLIs and SDKs
- Demos and examples
- Weird experiments that might become real features
Some repos are stable. Some are spicy. All are moving.
- Primitives over platforms
- Interop over lock-in
- Fast by default
- If itās not simple, itās not done
Agent Relay should fade into the background and just work.
Early. Actively built. Opinions welcome.
APIs will change. Thatās the fun part.
- Docs: https://docs.agent-relay.com
- Hosted Version: https://agent-relay.com
Built for agents that shouldnāt have to work alone.