Experimental · Windows · macOS · Linux

Stop being the clipboard between your assistants.

Claude Fable 5.2 ↔ GPT-6 Astra. You stop carrying the messages.

  • Uses the subscriptions you already have. No API keys.
  • Works in the Anthropic and OpenAI desktop apps.
  • Calls show up as a real chat in the other desktop app.

Before and after

Getting started

Install both desktop apps, then the bridge from npm.

  1. Install Claude Desktop

    Download the desktop app and sign in.

    Download Claude Desktop
  2. Install ChatGPT Desktop

    Install the desktop app, sign in, and open Codex.

    Get ChatGPT Desktop
  3. Install the bridge

    One npm command, or clone the source from GitHub.

    See the commands

From npm

Published · 0.1.0
npm install --global nomoremeatproxy@0.1.0
nomeat --help
nomeat orient --from astra

# one-off, without installing
npx --yes nomoremeatproxy@0.1.0 --help

--help and orient only read. No message is sent.

Package on npm

From GitHub

Public · MIT
git clone https://github.com/KanuToCL/nomoremeatproxy.git
cd nomoremeatproxy
node bin/nomeat.mjs --help
npm run doctor
node bin/nomeat.mjs orient --from astra

The last two only check and read. No message is sent.

Repository on GitHub

Send a test call

nomeat peers
nomeat call --to <peer> --from <caller> --message "Reply with one word."

Replace <peer> and <caller> with names from the first command. A reply means the message got through. From a source checkout, write node bin/nomeat.mjs for nomeat.

Tell both assistants

Add the section in templates/peer-calls.md, shipped in the package, to the instruction file each assistant reads: AGENTS.md, and CLAUDE.md if it does not import it. Installing does not edit them.

The bridge does not install Node.js or either command line, and signs nobody in. Peer names and models are fixed in the code for now. Bridge calls are tested on Windows, macOS and Linux. MIT licensed.

On your phone

Illustration · A chat open in its desktop app, with Remote Control on and active

Claude side only, through its Remote Control. The computer stays on with the chat open.

How it works

Basic

You ask either one
GPT-6 Astra
the bridge
Claude Fable 5.2
  1. Ask either assistant.
  2. It calls the other through the bridge.
  3. The other one answers.
  4. The reply lands in your chat.

Illustrated pairing. Peers and their models are fixed in the code for now.

Advanced

Calling assistant a Codex or Claude Code session

nomeat call --to <peer> --from <caller>

nomeat

Node.js 22.15+, no dependencies

Resolve peer

name → vendor + model (src/peers.mjs)

Mark the message

[nomeat call from <caller>; request <id>; hop n/m; AI, not user]

Ledger

ids, states, hashes, sizes. Never message text.

Guards

new sessions read-only by default · repeat-request checks · hop limit 4 by default

Codex peer

through codex app-server; a new thread is then opened in the Codex app by its link

delivery: direct

queued when the app has the thread open; the app runs the turn

Claude peer, nobody listening

a Claude Code run outside the app, then listed in the Claude app by its desktop opener, on Windows and macOS

delivery: headless

Claude peer, a chat is listening

mailbox → the open chat prints the request, answers, and returns it with nomeat reply

delivery: attended

one JSON result: state, threadId, reply, delivery

Supported entry points only. Signed-in vendor CLIs. No API keys.

Many machines

You one seat, any device

Windows 11

works today

GPT-6 Astra bridge Fable 5.2

macOS

works today

GPT-6 Astra bridge Fable 5.2

Linux

works today

GPT-6 Astra bridge Fable 5.2

  1. Put the pair on each machine.
  2. Steer every chat from one seat.
  3. Each pair works through its own bridge.
  4. Work moves between machines through the desktop apps.

One subscription per vendor. Chats persist. Machines do not call each other yet.

Remote view comes from each app's own remote access.

Status

Works today

On Windows 11, macOS and Linux.

  • Calls both ways, reply returned
  • Calls show up as real chats in both desktop apps
  • One continued conversation
  • Model and effort per call, where the app applies them
  • A listening Claude chat answers itself
  • Send without waiting and check later, Codex side
  • One side builds, the other reviews
  • Follow along from a phone, Claude side
  • Chat lookup by title, folder, or text in recent chats

Known limits

Observed in testing as of .

  • A question card in the first message of a new bridge-made Codex chat does not reach you. A card in a follow-up, once the chat is up, does. The bridge tells the assistant to put its question in the answer instead.
  • On Linux, a bridge chat is not listed in the Claude app. The vendor's opener exists for macOS and Windows only. On Linux the conversation shows when a Claude chat there is listening.

Why another bridge?

Others connect the two vendors too. This one continues real chats in both desktop apps, both ways, through the vendors' own command lines and the subscriptions you already pay for. On the Claude side the chat has to be listening.

  • Agent Tincan Web and local chats from many vendors, through a relay The closest design. It sends by driving the signed-in web pages, needs a Tailscale network, and its relay does not run on Windows.
  • CrossTalk One Claude desktop conversation and one ChatGPT browser tab Windows only. Links one pair at a time and copies each reply across as plain text.
  • Duplex Bridge A Claude chat to Codex, a ChatGPT chat to Claude Starts a fresh run for every call instead of continuing a chat you have open.
  • Frenemy Claude Code and the Codex command line call each other The smallest working example. A fresh run for every call.
  • codex-plugin-cc Claude Code hands work to Codex Official and supported. One direction, for review and delegation.
  • ChatSearch Search across your ChatGPT and Claude web chats Solves a different problem. An assistant cannot call it.

Read from each project's docs and source on 10 November 2026. None was installed or run here.