I found an interesting extension for VSCode that allows you to use a local whisper server to transcribe your thoughts into cursor chat commands
Launch a local whisper server
cd ~/dev
git clone https://github.com/martin-opensky/whisper-assistant-vscode
cd whisper-assistant-vscode
docker build -t whisper-assistant-server .
docker run --restart unless-stopped -d -p 4444:4444 --name whisper-assistant whisper-assistant-server
There is a missing module openai in the extension startup
but now I have a local instance of the whisper transcribe api
solution was to npm install in the install folder ~/.cursor/extensions/martinopensky.whisper-assistant-1.1.0
https://github.com/martin-opensky/whisper-assistant-vscode/issues/12
Last, modify the shortcut "Whisper: Toogle Recording" shortcut to use alt+cmd+\ instead of cmd+m (otherwise it won't record in the cursor chat).
Bonus: the whisper server hosts a openai compatible stt API, so you can re-use this everywhere.