At Summation, we use different coding agents for different jobs. Claude might work on an API while Codex reviews a change. But we kept doing the same manual work between them: copying requests, findings, and replies from one terminal to another.

We built Agent Exchange (AX) so those agents could communicate inside the terminals and conversations we already use.

Keep your terminals and let the agents talk

Give each agent a name, and it can ask questions, hand off work and reply to other agents without a person carrying every message.

Two agents start from scratch: Claude asks Codex for a review, Codex spots a bug, Claude fixes it, and Codex takes another look. We trimmed the setup and wait time.

Agents can review, respond and iterate directly while staying in their own sessions and keeping their context. AX works with Claude Code, Codex CLI, Grok Build, OpenCode and Pi.

How it works

Launch each agent through AX and give it a name. One terminal each:

In the OpenCode terminal, ask:

OpenCode gets the details from Claude, builds the UI and sends its changes to Grok for review. Replies return to the same sessions, so the agents can keep working from the context they already have.

We built AX around the things that mattered in our own day-to-day work:

• It stays on your machine. AX’s broker and mailbox run locally. There’s no cloud relay or telemetry. Your agents still talk to their own AI providers, as they always do.

• Your permissions still apply. Each agent keeps its own sandbox and approval rules. A task doesn’t gain extra permissions because another agent requested it.

• It works across repos. Your API and frontend can live in different folders. Agent names work across all of them.

• You can pick up where you left off. Launch an existing conversation through AX using the harness’s normal resume syntax. Relaunching the same AX name restores its saved conversation.

• Handoffs are visible. Run ax inbox to see messages and their delivery status. Acknowledgment confirms receipt, not completion of the task.

Why AX is open source

AX grew out of our day-to-day engineering work at Summation. We’re sharing it so other teams can use it, adapt it to their workflows and help improve it. It’s free and open source under the MIT license.

Try it

Install in one line on macOS, Linux or WSL 2:

Or let your agent do it. Paste this in:

Docs and quickstart at useax.dev. Code on GitHub.

Try it with the agents you already use, and tell us where coordinating them still gets painful.