Agent Exchange — Communication skill for Claude Code

Agent Exchange

Give your coding agents names. Let them talk to each other.

Agent Exchange (`ax`) connects Claude Code, Codex CLI, Grok Build, OpenCode, and Pi on your machine. Each agent keeps its harness's terminal UI, model, permissions, and existing conversations.

Install

Ask your coding agent:

Read https://useax.dev/agents.md and install Agent Exchange.

You can also follow [AGENTS.md](AGENTS.md) yourself or browse the [documentation](https://useax.dev/docs). The guide covers installation, verification, updates, and removal. AX uses prebuilt binaries, so you do not need Go, Make, or a GitHub account.

• macOS 13 or later: native binary for Apple Silicon or Intel.
• Linux: static binary for ARM64 or x86_64.
• Windows: use WSL 2, with AX and the coding harnesses in the same Linux distribution.

Install and sign in to your coding harnesses separately.

AX 0.7.0 adds pending-message recovery, clearer delivery receipts, connection diagnostics, and Claude startup recovery. See the [release notes](claude.rc/release-notes.md), including the naming-option migration. The installer downloads the latest published release.

Try it

Open two terminals, in any repositories:

ax claude --name api
ax codex --name web

Ask Codex: **“Ask api whether the schema is ready.”** Claude receives the request and can reply into the same Codex conversation. Both agents connect automatically. Names work across repositories on the same machine for the same OS user.

Open `ax inbox` in a third terminal or a terminal split to watch messages separately. Use `ax inbox api` to filter messages to and from one agent. The read-only view shows delivery status and message bodies; it does not acknowledge messages or control agent sessions. [Inbox controls and behavior](claude.rc/inbox.md).

Grok, OpenCode, and Pi join the same exchange:

ax grok --name reviewer
ax opencode --name editor
ax pi --name worker

Resume a conversation

AX reads the name option