Chat and the live assistant

Minutist has a built-in chat agent you can ask about a meeting. It answers from the record (the transcript, your notes, and any attachments) and it runs locally, on the same model that writes summaries.

This is the app’s own agent. It is not the connector, which is a separate feature for pointing an external assistant (Claude, Codex) at your meetings.

Using chat

Turn on the Chat column from the pane toggle. Chat is scoped to the meeting you have open; you can keep more than one conversation per meeting and switch between them. Type a question and the reply streams back.

How it answers

Rather than reading everything, the agent retrieves the passages relevant to your question (from the transcript and from attachments) using a local embedding model (semantic search combined with keyword search). It can also call read tools on the meeting: read a slice of the transcript, get the summary or notes, report per-speaker talk time, re-listen to a section, or re-summarise a different way. It works on the open meeting and never leaves your machine.

The live assistant

The live assistant is the chat model working while you record. Turn it on under Settings → Processing → Live assistant:

Off (default): no live agent.
Auto: run it only on a discrete GPU, where it will not compete with transcription for memory.
On: run it on any active GPU.

When on, it reads the transcript as the meeting runs and keeps a running digest: action items, decisions, and open questions, in a panel that updates as things develop, so you have live context to ask about.

During and after

After a meeting, chat has the full picture (transcript, notes, attachments, and the summary) and every tool available. During a recording, the live assistant is the piece that runs, gated on your GPU as above.