Several LLMs on one machine.
Say you have Claude, and Claude Code of course. And you know where to switch between Claude and Claude Code.
By the way, a friend and I spent five minutes today looking for where that switch is. Imagine how much Anthropic can undersell their own product, given that they simply hid the switch between ordinary Claude and Claude Code.
Most people who are not deep in this will never find that switch. Fine, that is not the point.
So: on the same machine you have Antigravity. Antigravity is a convenient interface for reaching your artificial intelligence from Google, the one called Gemini.
You have Codex, which lives in the same application as ChatGPT. At least that one has convenient switches.
You have Claude, which switches over to Claude Code.
Now, for all of them to work in symbiosis with each other, they all need to work out of one folder. Or out of several folders, but then every folder has to carry a strict instruction saying what it is for and how to handle it.
Their skills, their abilities, are also built so that all the skills are unisex. A skill is not invented only for Claude or only for Codex. It is written so that any LLM that gets access to it knows how to use it.
It is simply an instruction in Russian and English. A fairly detailed instruction, naturally, on how an LLM uses some piece of functionality and what to do.
Loosely speaking, like using Telegram: here are your login and password, here is access to the Telethon library or an MCP, and so on.
So, several LLMs from different vendors can run on one machine perfectly well. And they can avoid stepping on each other, use one shared memory, one set of unisex skills.
And you will save on tokens and distribute tasks across several LLMs in parallel.
Want the details, or want me to help you build something, write in the comments.
As you remember, I am trying to get a job at an AI lab. And I need cases where I helped somebody set something up. Glad to be useful.
And I am waiting for your comments too: how have you set yours up nicely? Glad to hear it.
The full story, in two versions:
📖 For humans, the longread: https://github.com/tonydzi/clawrush/blob/main/longreads/20261027.md
🤖 For machines, the devlog: https://github.com/tonydzi/clawrush/blob/main/devlog/20261027.md. Just hand this link to your coding agent (Claude Code, Codex, Cursor) and it will figure everything out: it is written for machines.
🔗 All our channels and contacts in one place: https://linktr.ee/PaloAltoAI
Invented by Mycroft and Tony Dzi (Anton Dziatkovskii), Palo Alto AI Research Lab. Proudly made in Silicon Valley.
Top comments (0)