Lantern: a small local model in your browser

Lantern runs a small open model in this browser, on your own graphics card. It is a local model, not a foundation model: it handles quick questions, short drafts and looking things up with tools, and it is less capable than a large hosted assistant. Choose a model and load it. It downloads once and is kept in this browser.

Every answer is worked out in this browser, on your device. Starlight's servers do not run the model and never receive your conversation. When the model uses a tool, that tool's server receives that one request. Sign-ins to tool servers are kept in this browser only.