AllMyAgents

AllMyAgents

**Run a whole fleet of coding agents from one window — and let them talk to each other.**

A self-hosted hub and desktop app for [Claude Code](https://claude.com/claude-code) and [OpenAI Codex](https://developers.openai.com/codex). Several accounts per vendor, side by side, on your own machine. Your agents can message each other, look in on each other's work, and share what they learn.

[](LICENSE) [](#install) [](https://github.com/nathanfraske/AllMyAgents/releases)

**Alpha.** It works, and it is used daily to build itself. Expect rough edges — read the release notes.

---

Why this exists

One agent in one terminal is easy. The trouble starts at five: which of them is still working, which quietly failed an hour ago, which two are about to edit the same file, and what did the one you closed yesterday actually learn?

AllMyAgents is the answer to that — a hub that owns the agents, keeps their history, and gives them a way to reach each other.

What it does

🤝 Agents that work together

A real intercom, not a metaphor. Any agent can message any other, and the hub delivers it into their next turn — or steers it into a turn already running, so a correction lands while it can still change the outcome.

• send_message — direct or broadcast to the fleet
• peek_agent — see what a teammate is doing without interrupting them
• Shared memory — scoped notes that outlive the chat that wrote them
• Practices — conventions the fleet reads and revises together

Mixed fleets are the point: a Claude agent and a Codex agent talk to each other the same way, and the transcript shows you which is which.

🪟 A window built for many agents at once

• Split any way you like