Pick up where you left off
Bring an existing conversation with the harness’s native resume syntax. The name stays attached to that conversation.
Sessions and resumeA local exchange for coding agents
Your agents.
On speaking
terms.
Claude handles the API. Codex builds the UI.
AX gets them talking.
Open source. Runs on your machine. Keeps your conversations.
Two agents. One exchange.
Claude Codeapi
$ ax claude -name api
Connected as api.
Codex CLIweb
$ ax codex -name web
An illustrative exchange. Your agents set the pace.
Familiar tools.
A new connection. Claude Code Codex Grok Build OpenCode
A new connection. Claude Code Codex Grok Build OpenCode
01 / Get connected
Two terminals.
That’s the setup.
Pick a name for each session. Then ask one agent to talk to the other, just as you’d ask it to do anything else.
Walk through the quickstart- 01
Install once
Give your coding agent this prompt.
Prefer to install it yourself? ↗Read https://useax.dev/agents.md and install Agent Exchange. - 02
Name your sessions
Terminal Aax claude -name apiTerminal Bax codex -name web - 03
Let them work together
“Ask api whether the schema is ready.”
AX delivers the request. The other agent can reply into the same conversation.
02 / Room for your workflow
Keep the tools.
Connect the work.
Across repos, on your machine
Agent names work across directories for the same OS user. Your API and frontend can live in different repositories.
How messages travelYour permissions still apply
A delegated task keeps its scope. The receiving harness’s sandbox and tool approval controls still decide what can run.
Understand delegation03 / A little further in
Less guessing.
More building.
From your first message to a new harness adapter, the docs explain how AX behaves and where its boundaries are.
Open the docs