Skip to main content
Gobi runs on AI models — that’s what turns a conversation into a transcript, a transcript into a Summary, and your questions into your agents’ answers. This page is about the other direction: what those models get from you. The short version — models work for you; they don’t learn from you.

​
Where models do their work

Three jobs, all of them yours:
  • Transcription. Turning a capture’s audio into its written transcript.
  • Summaries. Distilling the transcript into the Summary that lands in your Home.
  • Agents. Answering you — your personal agent from your Home, a space agent from what its space was given.
In every case the model is doing a job you initiated, on data you already own, with the result delivered back to you. A model call is processing, not publishing: it doesn’t share your conversation with anyone, and it respects every boundary these docs describe — a model working on your capture serves you, and only you.

​
We don’t train on your conversations

Gobi doesn’t use your captures, transcripts, Summaries, notes, or messages to train models — ours or anyone else’s. Your conversations make your Gobi smarter, through your agents; they don’t make the models smarter, and they never become anyone else’s answer. If that ever changes for some future feature, it will be opt-in, asked plainly, and off until you say otherwise. You won’t discover a training toggle you were supposed to find.

​
The models behind Gobi

Gobi uses leading AI models through their business platforms, under terms built for exactly this arrangement: providers process your data to serve your request — they don’t use it to train their models, and they don’t keep it beyond what processing requires. Model providers are infrastructure to Gobi, the way a server is: they do the work and hand it back. The full data picture — who can see what, and the controls you hold — lives in how Gobi handles your data and export or delete.