On-device embeddings
Optional local embedding pipeline (WASM + small model packaged with the browser) for semantic features without calling a remote model hub—explain privacy benefit and limitations.
What this is
Some builds ship a local embedding stack (WASM + small model) so semantic features can run without sending text to a remote model hub—improving privacy for supported flows.
How to use it
- No user setup is required when packaged; features activate when enabled in that build.
- Expect larger download sizes on first install when embedding assets ship.
- Quality differs from cloud-scale models; use cloud features when you need maximum recall.
Notes and limits
Embedding coverage does not replace full-page understanding for arbitrary PDFs.
Related topics
Related Documentation
Active tool indicator
Shows when a browser “tool” action is in progress so users know the assistant is acting on the browser, not only typing text.
Assistant themes
Multiple light and dark themes for the assistant chrome (not only system theme); choice syncs via prefs / bridge.
Busy / streaming state
Visual feedback while the model or tools are working (AssistantBusyBar, streaming into the timeline).
About the Author
I'm the CEO of Kahana, bringing a unique perspective from my management consulting experience at Clarkston Consulting and biomedical engineering background from Duke University. I'm focused on making the future of work more elegant through innovative technology solutions that prioritize user well-being and productivity.