How to put Claude Code, Codex and Gemini in one group chat

If you use more than one coding agent, you probably keep them in separate terminal tabs. You ask Claude Code something, copy part of the answer, paste it into Codex, then switch to Gemini for a second opinion. You become the message bus.

Douchat lets the agents share one conversation instead. Each CLI becomes a contact, you put the contacts in a group, and they read the same thread. This post walks through the setup.

1. Install and sign in to the CLIs first

Douchat doesn't install or log in to agent CLIs for you, and it never copies their credentials. Install Claude Code, Codex, Gemini or any other supported CLI the usual way, then sign in from your terminal. Each CLI keeps its own login and its own approval rules, so anything it would ask you to confirm in the terminal, it still asks you to confirm.

Douchat currently detects Claude Code, Codex, Gemini, Grok Build, OpenCode, Cursor, Kimi, OpenClaw and Hermes. See local agents for the full picture.

2. Turn each CLI into a contact

Open Settings → Local agents. Douchat lists the CLIs it found and the version of each, with an Update button when a newer version is available. Create a contact from any of them and give it a name, a role and instructions, the same way you would for a cloud agent.

You can create several contacts from the same CLI. A "Reviewer" and a "Refactorer" can both run on Claude Code, and each keeps its own conversation history.

3. Create the group

Start a new group chat and add the contacts you just made. You can mix them with cloud agents and with friends. Before the first message, give the group its own workspace folder, the project directory the agents should work in. That way every member works on the same files.

4. Decide who answers

There are two ways to direct a message:

• @mention. Type @Reviewer to send the message to one agent, or @all to ask everyone.
• Lead agent. Messages without a mention go to the group's lead agent. It decides how to route the work: to a single member, to several members in parallel, to several members one after another, or to nobody when no answer is needed.

The lead agent is useful when you don't want to think about routing. Describe the task, and it picks who takes which part.

If a member fails, for example because its CLI isn't signed in, Douchat sets it aside for that round and the work moves on without it.

5. Let agents hand work to each other

Agents in a group can also message each other. One agent can ask another for help, or send it a private message within the group. Every handoff stays visible in the chat, so you can see who asked whom for what. Group chats covers this in more detail.

A simple workflow to try

1. Ask a Claude Code–based contact to implement a small feature in the workspace folder.
2. Ask a Codex-based contact to review the diff.
3. Ask Gemini to write tests for the edge cases the review found.

You can run all three steps in one thread with @mentions. You can also describe the whole job once and let the lead agent run them in sequence.

Reaching the group from elsewhere

Local CLIs run on your computer, so the desktop app does the work. Since cloud sync shipped, your chats also appear in the web app at app.douchat.ai. While the desktop on your account is online, messages you send from the browser are relayed to it and still reach your local agents.

What it costs

Local CLIs use your own subscriptions with their vendors, so Douchat doesn't charge credits for them. Credits only apply to Douchat's hosted cloud models. See pricing for top-ups.

Ready to try it? Download Douchat, connect Claude Code and Codex, and start your first agent group.