I use Codex, Claude Code and Gemini CLI on the same projects. The worst moment is when one of them hits its usage limit halfway through a feature. The next AI starts cold: it re-reads half the project, asks what we are building, and still misses the decisions I already made.
After a lot of trial and error, I made which-ai-to-use: a free, open-source tool that hands your task to another AI with the full context, so it carries on instead of starting over.
Try it: install it from VS Code (search which-ai-to-use in Extensions), or run npx which-ai-to-use in any project.
What it does
-
Hands your task to another AI. When the limit hits, it writes a handoff note (
HANDOFF.md) with the task, your latest requests, where the AI stopped, the commits and changed files, and your test command. Codex, Gemini CLI or any other AI reads it and carries on instead of starting over. - Reads the chat for you. The task and your requests come straight from your Claude Code, Codex or Gemini CLI chat, so you don't retype anything.
- Keeps your secrets out. API keys, passwords and tokens from the chat are hidden before anything goes into the note.
- Picks who carries on. It suggests the best AI you have installed for the task and starts it with the note. No other AI installed? It tells you which one to get.
- Gives the task back. When Claude's limit resets, it offers to hand the task back to Claude with the same note.
- Tells you which AI to use. For any task, it picks the best of your AI tools and the right model, with a ready command.
-
Helps you hit the limit less. It warns you before your 5-hour window runs out, with ways to make it last (like
/compactor a smaller model for small edits), and makes your AI skip files that waste tokens: onepackage-lock.jsoncan be 50,000+ tokens. - Shows your AI week. Tokens per day next to the week before, and requests per AI.
How to use it
- In VS Code, Cursor or Windsurf: open the Which AI sidebar and click Claude limit over?. It writes the note and shows who should carry on, with a Run in terminal button.
-
In a terminal: run
npx which-ai-to-use handoffin your project. - Automatically: the optional Claude Code plugin writes the note the moment the limit is reached. Setup is in the README.
Get it
- VS Code: Which AI to Use on the VS Code Marketplace, or search which-ai-to-use in Extensions
- Cursor, Windsurf, VSCodium: Which AI to Use on Open VSX
-
Any terminal, no install:
npx which-ai-to-use - Source: github.com/WST2002/which-ai-to-use
Everything runs on your laptop and nothing is uploaded. Usage warnings and reports read only numbers from the logs; the handoff reads your latest chat, on your laptop, to write the note. It is tested on real projects (199 tests).
What do you put in your handoff notes? I'd love to know what I'm missing.
