# How to Switch AI Models Without Losing Anything

A practical system for switching between AI models in hours without losing your history, skills, or project context. It combines a shared conversation archive, synchronized Skills, and portable project profiles.

Author: Maxim Tolmachev
Published: 2026-11-13
Language: en
Canonical: https://maximtolmachev.com/blog/how-to-switch-ai-models-without-losing-anything/
Original source: https://t.me/mtolm/109

![How to Switch AI Models Without Losing Anything](/blog-images/tg-109.jpg)

How to switch from one AI model to another in a couple of hours without losing anything

“Hi Maxim! You mentioned that you moved to Codex in a couple of hours. Are you sticking with it as your main platform, and do you have any tips for making the move quickly?”

Dima sent me this question after my post about editing a podcast with AI.

I’ve already covered which AI model is currently the best value for different tasks in my posts about choosing between Claude and Codex for $200, the model atlas, and the LLM economics map. Here, I’ll explain it in really simple terms, without complicated technical jargon: how to set up your work with AI so you can switch models like gloves and never depend on a single subscription.

At the end of the post, I’ve included a ready-made prompt in Google Docs. If you want to build the same system for yourself, you won’t have to dig through the settings manually—the AI will configure everything step by step 👇

1️⃣ A shared journal of every conversation (like an electronic medical record)

The biggest pain when switching AI models is the feeling of amnesia. It’s like seeing the same doctor at a private clinic for six months, then going to a new one who has a blank sheet of paper and making you retell your entire medical history from scratch.

It’s the same with AI models: you’ve spent months discussing tasks and making decisions in Claude, then you open ChatGPT or Codex—and it’s seeing you for the first time.

To prevent this, I have a free open-source program called Retain installed on my computer. I published my own improved one-click installation version on GitHub.

It works like a single medical record: in the background, it automatically downloads all my conversations from every chat and app I use—Claude, ChatGPT, Codex, Cursor, Gemini, and Antigravity—and syncs them with my cloud server every 15 minutes. The archive now contains more than 7,660 conversations and 141,000 messages.

As a result, whenever I open any new AI model, I don’t need to retell six months of work. I simply write: “Check the Retain archive and see what we agreed on for project X.” It finds the old chats in a second and immediately continues the conversation from the right place.

2️⃣ One box containing 153 skills (like a chef’s recipe book)

A bare AI model out of the box is like an intern on their first day: smart, but doing everything their own way. To make it useful, you need to teach it your “recipes” (skills / Skills): how to write in your style, check spreadsheets, prepare résumés, analyze calls, or compile analytics.

People usually train one AI model inside its own window, then have to teach the new “intern” from scratch when they move to another app.

All 153 of my skills are stored in one main folder on my computer, like a recipe book. A special background courier script automatically distributes copies of these skills to every AI app on my laptop (7 apps) and to my cloud server (another 10 environments).

Teach something once, and all my AI models gain that ability the same minute—even if my laptop is closed and I’m working from my phone. I’ve also open-sourced my skill templates and the synchronizer itself on GitHub: the Skills templates and daily-skill-sync synchronizer.

3️⃣ An “owner’s passport” and a guide to 17 projects

Even if an AI model has a conversation archive and a box of skills, when you first meet it still needs to quickly understand the basic rules of the game: who you are, what you do, which projects are currently active, and what tone it should use with you.

Trying to feed every AI model the exact same settings file directly is a bad idea: different apps have their own formats and conventions.

That’s why I have a single “passport” folder containing a short table-of-contents-style guide, a description of my profile, and separate cards for each of my 17 projects. A small translator script automatically turns this folder into a clear starting cheat sheet in the “native language” of each specific app.

Whenever a new model comes out—whether it’s American or a Chinese model like Qwen or DeepSeek—I write one sentence in my very first message: “Read the starting card and the guide.” Ten seconds later, it knows all my rules, projects, and context as well as an assistant I’ve worked with for a year.

🎁 How to build the same system for yourself (without programming)

When you own your chat history, your skills sync automatically, and your project cards are translated for any app, moving to a new AI model takes a couple of hours instead of weeks. You’re no longer tied to one corporation—you simply use whichever model solves today’s task better and more cheaply.

So you don’t have to configure all of this manually, I packaged detailed step-by-step instructions for an AI agent into a separate document:
👉 Get the prompt for setting up the same AI system (Google Doc)

Just copy the text from the document into Claude Code, Codex, Cursor, or Antigravity. The agent will ask you four simple questions about your tasks and build the entire system around your needs.

P.S. I hope the pictures in this post made you laugh too 😄 I asked Gemini, which helped me put the article together, to delegate the illustration generation to a Codex agent—but I also told it to use Claude’s frontend-design skill, which makes it build everything in HTML. So instead of generating ordinary images, Codex literally drew all these doctors, chefs, and robots with HTML/SVG code directly in the browser and took screenshots. That’s why they turned out so unhinged and funny. I decided not to redo them and just leave them as they are!
