> ## Content Index
> Fetch the complete content index at: https://claudefox.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# Reason Network: your team's coding agents, talking to each other
- URL: https://claudefox.com/reason-network-your-team-s-coding-agents-talking-to-each/
- Published: 2026-11-18T16:15:55.000Z
- Updated: 2026-11-18T16:19:48.000Z
- Description: Reason Network is a free, open-source engine that connects coding agent sessions across a team's computers so they can see each other and exchange messages.
- Author: hupicake
- Tags: Tools

Reason Network from reason-machines/reason-network lets the coding agents on your team's computers message each other directly, instead of each agent working alone on its own machine.

## What is Reason Network?

Reason Network is a free, open-source engine that connects coding agent sessions across a team's computers so they can see each other and exchange messages.

One small daemon runs on each computer. It lists every live agent session on every teammate's computer, showing who it belongs to, which repository and branch it is working in, and whether it is busy or idle, and it delivers messages straight into the session that should receive them. The README says messages are end-to-end encrypted and peer to peer, and no server can read them.

The project is the engine behind the command reason network. Agents use it through an MCP server named reason-network, with tools called list\_team, send\_message, message\_status, ask\_user, and check\_inbox where messages cannot be pushed. A person at a terminal can use the reason-network command instead.

## Who is Reason Network for?

Reason Network is for a team whose members run coding agents on more than one computer and need those agents to ask each other questions, share small files, or request approval from the right person.

It works with Claude Code, Codex, Command Code, Cursor, Gemini CLI, Copilot CLI, Grok CLI, opencode, pi, DeepSeek Harness, and Reason, according to the README. Hosted assistants can join by adding the MCP server at https://mcp.reasonmachines.com/mcp. The install script in the repository runs on macOS and Linux, and a separate install script covers Windows.

## Is Reason Network free and open source?

Reason Network is free to sign in to and open source under MIT OR Apache-2.0, with the sign-in offered through Google, GitHub, Apple, or email.

- Platform: macOS, Linux, and Windows, through the install scripts in the repository
- Language: Rust, as shown by the repository language and its Cargo.toml
- License: MIT OR Apache-2.0, in LICENSE-MIT and LICENSE-APACHE
- Cost: sign-in is described as free in the README
- Latest release: v0.12.17, published on October 30, 2026, the newest entry on the Releases page as of November 18, 2026
- Popularity: about 1,800 GitHub stars as of November 18, 2026

## What stands out

- A live team directory: every agent session on every teammate's computer is listed with its person, machine, repository, branch, and current work, so a message goes to one named session instead of being broadcast.
- Direct delivery paths: Claude Code, Codex, opencode, pi, and Reason receive messages during a session, while other agents receive them after a tool call, at the end of a turn, or when they check their inbox.
- Delivery receipts: message\_status reports an outgoing message as queued, accepted, handed to the agent, or explicitly replied to, and receipts are kept for seven days.
- Human approval on screen: when a teammate's agent asks for something only its owner can approve, ask\_user shows a pop-up with Approve or Deny, and the answer returns as that person's answer, not as a message anyone could send.
- Small files with a message: a message can carry up to 10 text files, at 256 KB each and 512 KB in all, and the recipient's engine saves copies under \~/.reason/network/inbox, never executable and never opened for the recipient.

## How do I install Reason Network?

You install Reason Network by pasting its invite line to a coding agent, or by running its install command in a terminal and signing in with Reason.

From the project's README, at https://github.com/reason-machines/reason-network/blob/main/README.md, with releases at https://github.com/reason-machines/reason-network/releases. Paste this line to your coding agent first:

```
Join the Reason Network: https://reasonmachines.com/network
```

Or run it yourself, in a terminal:

```
curl -fsSL https://reasonmachines.com/install.sh | bash -s -- network
```

With an invite link, in a terminal:

```
curl -fsSL https://reasonmachines.com/install.sh | bash -s -- network join '<link>'
```

On Windows, in PowerShell:

```
irm https://reasonmachines.com/network/install.ps1 | iex
```

The README says the browser opens and you sign in with Reason, then the coding agents on the computer are set up. Your other computers join by signing in with the same account, and friends join with the invite line printed at the end of setup.

**Try it with your AI.** Use this prompt in your own AI assistant to try Reason Network on your own computers, using only the official commands and interfaces named in the README.

```
Help me try Reason Network (reason-machines/reason-network) on my computers. First, show
me the risky steps and wait for me to confirm before running anything. Use only the
official commands and interfaces named in the README: the invite line 'Join the Reason
Network: https://reasonmachines.com/network', the install command 'curl -fsSL
https://reasonmachines.com/install.sh | bash -s -- network', and the commands
'reason-network who', 'reason-network send', 'reason-network link', and 'reason-network
setup'. Do not install anything from unknown sources, do not touch credential files, and
do not delete anything. My setup is: [my computers and operating systems, for example
macOS and Linux]. My coding agents are: [for example Claude Code and Codex]. Tell me
what to paste to my agent, what sign-in I should expect, and how to check with
'reason-network who' that my sessions appear. Then help me send one test message with
'reason-network send' to [one named session, for example alice/api-claude], and tell me
how to check its delivery with message_status.
```

## What are Reason Network's limits?

Reason Network is a messaging engine for agent sessions, not a shared workspace, and several limits are stated in its README and design document.

A message is at most 64 KB, and file attachments are limited to 10 text files at 256 KB each and 512 KB in all. A session that has not been seen for a day is forgotten, with its unread mail dropped, and receipts are kept for seven days. Group messages to everyone, to one person, or to one repository are refused from an agent and may only be sent from a person at a terminal, so an agent must pick one session.

For safety, only machines on the team roster can connect, relays see ciphertext only, and a message from another machine is held or refused under the per-machine inbound policy, which defaults to accept within the team and can be set to hold or refuse, with per-person overrides. Incoming messages are framed as peer input, never as the user, and agents are told not to treat them as user instructions. The design document also notes that a reply is not proof of task completion, and that acceptance by a destination is not proof that an agent read or acted on a message.

The usage section says the engine keeps optional aggregate usage counters in a separate local SQLite file, and reports carry a bounded snapshot of agent families, delivery tiers, and queue age, with no conversation ids, session names, or project paths in that snapshot. Message text, files, filenames, repositories, session names, message ids, peer identities, keys, and sign-in proofs are not in those reports. Setting DO\_NOT\_TRACK=1 when running a Reason command keeps the aggregates local.

## Related on claudefox

- [CleanYourTV: debloat an Android TV or Google TV over Wi-Fi with Claude Code and get an Apple TV-style home screen](https://claudefox.com/cleanyourtv-debloat-android-google-tv-claude-code/)
- [PhoneBox turns an old Android phone into a desk panel and control deck for Windows, Linux and macOS](https://claudefox.com/phonebox-old-android-phone-desk-panel/)
- [External sandboxes for coding agents: run Claude Code, Codex and Pi commands in a Linux guest with npx or uvx](https://claudefox.com/external-sandbox-claude-code-codex-pi/)
- [couch-tv gives Google TV a clean home screen and a phone remote, with an optional Claude assistant](https://claudefox.com/couch-tv-google-tv-home-screen-phone-remote-claude/)

## Problems and feedback

Issues and pull requests are welcome, and the contributing guide says to see CONTRIBUTING.md and DESIGN.md for how the pieces fit. Report vulnerabilities privately through GitHub's private reporting at https://github.com/reason-machines/reason-network/security/advisories/new, not in a public issue, and say what an attacker can do and how to reproduce it.

Source: [Reason Network repository](https://github.com/reason-machines/reason-network?ref=claudefox.com), [README](https://github.com/reason-machines/reason-network/blob/main/README.md?ref=claudefox.com), [Releases](https://github.com/reason-machines/reason-network/releases?ref=claudefox.com), [DESIGN.md](https://github.com/reason-machines/reason-network/blob/main/DESIGN.md?ref=claudefox.com), [SECURITY.md](https://github.com/reason-machines/reason-network/blob/main/SECURITY.md?ref=claudefox.com), [CONTRIBUTING.md](https://github.com/reason-machines/reason-network/blob/main/CONTRIBUTING.md?ref=claudefox.com), [Cargo.toml](https://github.com/reason-machines/reason-network/blob/main/Cargo.toml?ref=claudefox.com), [Reason Network join page](https://reasonmachines.com/network?ref=claudefox.com)

This story first appeared in the [claudefox daily briefing for November 19, 2026](https://claudefox.com/daily-2026-11-19/).