Getting started

This guide covers installing Minutist, completing the one-time model download, and recording your first meeting. All steps after the model download happen entirely on your machine: no account, no server, no telemetry.

1. Install the app

Choose your platform and follow the install guide.

Windows

See Install on Windows.

macOS

See Install on macOS.

Linux

See Install on Linux.

2. First run: download the on-device model

When you open Minutist for the first time, the app asks you to download the models it needs for transcription, speaker labelling, and summarisation. This is the only network operation the free build makes. You start it; nothing is sent out.

1. Open Minutist. The model download prompt appears.
2. Click Download models. The files are fetched once and stored locally.
3. Wait for the progress bar to finish. The app is ready to record when the indicator clears.

Note

Model files are stored in your local data directory. Sizes and exact filenames are shown inside the app during download; they are not listed here because they change between releases.

After this step, you can disconnect from the network. The entire pipeline (transcription, speaker labels, summaries, chat) runs locally from here on.

3. Pick your microphone

On first run (or via Settings → Capture), select the microphone you want Minutist to record from.

• macOS / Linux: the system microphone picker lists your available inputs.
• Windows: select your preferred input device; if you want to capture system audio alongside the mic, the system loopback option appears when your audio driver supports it.

Minutist asks for microphone permission the first time you start a recording. Grant it at the system prompt.

4. Take your first meeting

1. Click New meeting (or press the keyboard shortcut shown in the toolbar).
2. Enter a title, or leave it blank; you can edit it later.
3. Click Record. The transport bar turns active and the live transcript starts building on the right side of the screen.
4. Speak, or run the meeting as normal. Speaker identification (working out who spoke when) is on by default (the Identify speakers setting) and distinct voices are labelled as they are recognised.
5. Type notes in the left panel at any time. Your notes are stored alongside the transcript as you go.
6. Click Stop when the meeting ends.

Tip

The window is a row of columns (Notes, Transcript, Summary, Chat, and Attachments) that you show or hide with the pane toggle at the top. Notes is shown by default (and Summary, once a meeting is finished); the others are one click away. Each has its own page under Using Minutist in the sidebar.

The recording and transcript are saved to your local meeting store immediately.

5. Summarise locally

After stopping, a summary is drafted automatically; read it in the summary panel. You can press Summarise (it becomes Re-summarise) to run it again, edit the draft directly, or change the preset. It runs on your machine; no request is sent out.

The default summary is a short overview, the key decisions, and the action items. Presets for by-topic detail, or for skipping greetings and chit-chat, are in the summary panel.

The source recording, the transcript, and any notes remain available on the Meeting page.

More you can do locally

These all run on your machine, no request sent out:

• Attach documents and images. Drag files onto a meeting’s attachments pane or drop them straight into the notes editor and they render inline: an image appears as a thumbnail you can open full-size, other files as a card that opens in your OS app. Supported types include txt, md, csv, tsv, json, yaml, xml, log, html, eml, pdf, docx, pptx, xlsx (and ods), plus images (png, jpg, tiff). Image attachments are read by OCR using the local vision model, no extra download. Attached files feed the summary and on-demand retrieval.
• Retrieval from long attachments and the transcript. Ask about a 40-page attachment and the app retrieves the relevant passages rather than feeding the whole file to the model. The transcript is indexed the same way. Everything runs locally; the text-embedding model (~600 MB) is an on-demand download you start, the same class as the existing model downloads.
• Play a transcript line. Hover over any transcript row and a play control appears; click it to hear the original audio for that segment. Available on a saved, finished meeting.
• Live assistant (opt-in): on a machine with a discrete GPU, the live assistant 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. It is off by default; Auto engages it only on a discrete GPU, where it won’t compete with transcription for memory. The conversation carries on after you stop. See Chat and the live assistant.
• Name speakers once. Give a voice a name and Minutist recognises it in later meetings (a voiceprint). You can merge, rename, or forget identities.
• Translate. Translate a meeting’s transcript and summary.
• Reprocess a recording. Re-run transcription and speaker labelling on an existing recording in one action.

Where to go next

• Per-OS install details: Windows · macOS · Linux
• Phone companion (Android, connected tier): Phone companion records audio on the go and hands it to a signed-in desktop for processing over the encrypted sync channel.
• Connect an assistant: Connect Claude or Codex. The connector is free and local: point your own Claude or Codex client at the app’s MCP endpoint and it reads meetings you choose to share. Note that the connector sends the meeting content to the AI provider by design; see Security for the full breakdown of where your data goes.