Local LLM

Provides an on-device language model to websites the user approves.

Local LLM lets a website you approve use a language model through the standard window.LanguageModel API, the same shape as Chrome's built-in Prompt API. What you control • You choose which sites can use it. Every other site cannot see the extension. • The first time a site creates a session, you pick Always allow, Allow once, or Deny. • The default model runs on your device. Prompts for that model stay on your computer. • Cloud providers stay off until you turn one on and paste your own API key. The consent dialog tells you when a prompt will leave the device. You can use a supported cloud AI provider or a server you run yourself. • You can set limits and revoke a site at any time. The usage log records the site, the time, and token counts. It never records the prompt. Using it • Click the toolbar icon on a site and choose Enable on this site, then reload the page. • Or open the playground to chat without a website, attach images, and optionally turn on tools that act on your own tabs. • Download a local model from the options page, or let it download the first time a site uses it. For the site • The site calls LanguageModel.create() and session.prompt(). It receives the model's text. It does not receive your API keys or your other tabs. Local LLM has no ads and does not send your prompts to the developer. Chrome 116 or later. Local models are fastest with WebGPU.