Run AI on your own computer (Ollama, LM Studio)
Prefer that no request ever leaves your machine? BGP can use a model running locally. It's free and private. It needs a reasonably capable computer and an initial model download.
What you need
- Ollama (ollama.com/download) or LM Studio. Both install without admin rights.
- A model downloaded once (a few GB). A computer with 16 GB of memory or more is comfortable.
- The app running whenever you want BGP to use it.
Set it up
- Install Ollama and run a model once, for example
ollama run llama3.2. - Let Ollama accept browser connections: set the environment variable
OLLAMA_ORIGINS=*once and restart Ollama. LM Studio (usuallyhttp://localhost:1234/v1) works without this step. - In BGP, open Settings → AI Engine, set Default AI provider to On my computer,
open Advanced — connect your own local server, enter the server address (for Ollama
http://localhost:11434) and the model name, and click Test & save. Chrome asks once for permission to reach localhost.
Good to know
- The server address stays on this device and never syncs.
- For semantic memory search on your computer, pull the
nomic-embed-textmodel in Ollama, then turn on Settings → AI Engine → Semantic memory search → On your computer. - You can keep a cloud engine as well and switch in Settings → AI Engine any time.
Was this helpful?