A real-time desktop assistant that listens, reads the screen, and answers from your own documents — running locally by default.
Generic AI answers are useless when they do not know your material, arrive too late, or require sending private audio to a third party.
- Transcribes speech live and classifies what is being asked
- Reads the screen with a vision model, so it can answer about something never said aloud
- Retrieves relevant sections of your own documents through a local vector store
- Keeps recent context so follow-ups stay coherent
- Stores everything locally, with offline speech recognition
Provider abstraction over a hard dependency
Speech recognition swaps between local and cloud providers, and inference between hosted and local models, behind one interface — so a client with data-residency requirements gets the offline path as a config change, not a rewrite.
A correction pass after transcription
It fixes mishearings of domain vocabulary, which is where raw speech recognition consistently fails.
Model loaded once, not per request
Load time dominates inference at real-time latency, so the model stays warm.
If you need a real-time assistant over your own data, with privacy constraints that rule out sending everything to a cloud API.
Need something like this?
Tell us what you are trying to build. A few lines is enough to start, and you get a reply within one business day.