Phyrex|11月 13, 2026 09:09
ChatGPT + Grok + Claude 3-AI Collaboration Code Released
Without further ado, I was one of the first to release a multi-AI collaboration solution on X. After testing it for about a month, the code is now quite stable and allows the three AIs to work effectively together, so I'm releasing the code.
GitHub address: https://(github.com)/nisen0808-web/ai-collaboration-guide
Please consider giving it a star while you fork it
Please note:
1. This solution is scheduled locally; model inference is still handled by each provider's own service. You will need Codex, as well as Grok Build and Claude Code command-line clients (CLI). If you don't know how to install them, don't worry, you can provide the repository link to your Codex repository and let it assist with the installation and checks according to the configuration guide. Login and authorization are handled by you.
2. In this codebase, Codex is the team leader for the three AIs, responsible for task handling, task allocation, result checking, and delivery. Claude and Grok are responsible for analysis and review, respectively. The current public version is designed around Codex as the main controller; additional adaptation is needed to enable other AIs as team leaders.
3. Many developers like using bots, which is great. However, direct communication between bots is difficult, not impossible, but very costly. For example, my ChatGPT bots have been in use for a month, and there's a huge backlog of content. If an external bot were to access it, all the content would need to be fetched, and Grok's statistics show that each interaction requires millions of tokens.
Therefore, the best approach is to log into each other's pages on the cloud computers of both the ChatGPT and Grok bots. For example, logging into Grok on the ChatGPT bot's computer, and logging into ChatGPT on the Grok bot's computer.
This method allows for effective communication between the two bots, including file transfer.
4. Because execution is via CLI commands, each AI has certain limitations. If multiple tasks are running simultaneously, the other two AIs assisting in the execution may time out, which is unavoidable.
5. Currently, my bound system uses grok-4.7 with xhigh inference strength for Grok and Fable-5.1 with medium inference strength for Claude. These can be changed; you can communicate with the AI to change them if you encounter any issues.
6. The public version integrates collaboration tools with Codex via MCP. After installation, the service name starts with ai_team_ followed by a random number. The public version provides global collaboration rule templates, which can be merged into existing rules to allow subsequent chats to attempt collaboration when the necessary tools and permissions are available. You can disable the corresponding connection if you no longer wish to use it.
7. After completing this setup, I started preparing to integrate with Muse. Many people have asked me how to integrate it; I've already started working on it, and Muse also has bots, making the process even simpler. 8. Multi-AI Collaboration: Trust me, it's incredibly useful, especially in large projects. AI often veers off course, and it's crucial to guide it back to the right path. Regarding illusions, I haven't yet encountered two different AIs experiencing the same file simultaneously, especially since I configured three. However, multi-AI collaboration does slow down task execution, but the accuracy and success rate are incomparable.
9. You all know how to contact me. For any maintenance needs or to add new features, just let me know. AI collaboration may be the key to maximizing human productivity.
One @Gate, trading more markets.
Share To
Timeline
HotFlash
APP
X
Telegram
CopyLink