Stenograf help

Discover

Ask what was decided, who said what, or what changed. Discover answers from your saved transcripts and points the answer to the passages it came from.

Discover is in Alpha. An answer can only use the passages Discover found, and it can miss one, so read the quoted passages before relying on an answer.

Set up the local language model

Discover, summaries, translation and Live assistance share one language model that runs on your computer. In Settings → Local AI → Local language model, choose Built-in (Gemma 4, inside Stenograf, nothing else to install) or Ollama (a separate app on this computer), then Set up model. Nothing downloads until you choose to.

Choose Ollama if you want a larger, more capable model than the built-in one: any model Ollama keeps on this computer can be selected, as long as your computer has the memory to run it. Cloud models in Ollama are refused, because your transcripts must stay on this computer.

Ask

Type a question in Discover: what did we agree with the supplier about delivery dates? Discover searches all of your recordings and writes a short answer, pointing each line to the passages it came from and marking any line it cannot back up as unsupported. Each cited passage shows who said it; follow it to open that recording in History.

A Discover answer about a product launch date, each line followed by numbered citations, with the quoted passages and their speakers below.

To ask about one meeting, start from that recording in History. Follow-up questions continue the same conversation; when Discover restates a follow-up to search for it, it shows the question it actually searched. While Discover is open, your conversations are listed in the sidebar, newest first, with New conversation above them; Back at the top returns to the main menu.

Rating an answer

Rating is optional and is the one time transcript text leaves your computer. Before you send, Stenograf shows exactly what goes with the rating: your question, the answer, and the passages it quoted with their speaker and recording name. Nothing else from your recordings is included.