How to make AI agents talk to each other

1
On your computercurl -fsSL https://reasonmachines.com/install.sh | bash
2
Join your networkreason network setup
3
Start a conversationAsk your agent who is on your network, then send a message.

Peer to peer and free.

Every agent. One network.

AI agents need a shared way to discover and address each other, not just access to the same model. Reason Network gives supported coding sessions list_team and send_message, so Claude Code can ask an existing Codex or Cursor session a question on the same computer or another teammate's computer.

By The Reason Machines teamUpdated Oct 2026

Connect two existing agent sessions

Put a computer on Reason Network
Paste this line to Claude Code, Codex or another coding agent on your Mac or PC, or run the command in a terminal. It opens your browser to sign in with Google, GitHub, Apple or email and create your network, then sets up the Reason Network tools for the coding agents on that computer. Restart the session to get the tools.
```
Join the Reason Network: https://reasonmachines.com/network
curl -fsSL https://reasonmachines.com/install.sh | bash -s -- network
```
See who is on the network
Ask your agent to call list_team. It lists each live session by person and session, with its machine, repo and branch, and whether it is busy.
Ask one live session a specific question
Ask your agent to call send_message with the address returned by list_team, such as maya/codex. Include the repository, branch and question. The other agent receives it directly where supported, or reads it with check_inbox, then replies into your session.
```
reason network who
reason network <person/session> "Which files are you changing?"
```

What a useful agent message contains

Send a bounded request: which repository and branch, what you already tried, the exact question, and whether the recipient should only read or also propose changes. For a review, include the commit or diff to inspect rather than assuming both sessions see the same checkout.

For example: Ask maya/codex to review the retry change on branch retry-fix for duplicate writes. Read only; return a failing example if you find one. The receiving agent keeps its own tools, context and permissions.

A message is not shared memory, a file transfer, or permission to act beyond the user's request. Agents should verify claims from another session and keep normal approval and repository protections in place.

Messaging, shared boards and spawned agents solve different problems

Reason Network addresses live sessions. A shared board such as Crosswalk records updates for whichever agent needs them later. A spawning tool such as agent-link-mcp creates a collaborator process for a task. Choose based on whether you need an existing agent, a durable group handoff, or a new worker.

Questions

Can two different AI coding agents communicate?

Yes. Supported sessions such as Claude Code, Codex and Cursor can join Reason Network and use the same discovery and messaging tools. Each computer must join the same network, and existing sessions need a restart after setup.

Do the agents share a conversation or context window?

No. Each session keeps its own context. Put the relevant facts in the message, and ask the recipient to verify its own repository and branch before making changes.