Skip to content
Pre-launch. The free desktop build works today — it isn't code-signed yet, so your OS warns on first run. Paid sync, billing, and the mobile app aren't open yet. Get in touch.

FAQ

Free build: the only network operation is the model download you start yourself from the first-run prompt. After that, everything (transcription, speaker labelling, notes, summarisation, chat) runs locally. No account, no telemetry, nothing sent out.

Connected tier, sync: your meeting data is end-to-end encrypted before it leaves your machine. Our server brokers the connection and holds ciphertext for offline devices; it never sees your plaintext.

Connector (free build): when you ask Claude or Codex to read a meeting, your own assistant client reads it from the app’s local MCP endpoint and sends that meeting’s content to the AI provider. That is what the connector does; it is not a local-only operation, and we do not call it private. See Security for the full breakdown.


Is the connector channel encrypted or private?

Section titled “Is the connector channel encrypted or private?”

No. When you use the connector, the meeting you choose is sent to the AI provider (Claude, Codex) so the assistant can read it. The request goes over HTTPS, but the AI provider receives and processes the content. This is by design; there is no other way for the assistant to read your meeting. We are explicit about this because we never describe the connector as private or zero-knowledge.

Sync is different: it is encrypted so our server cannot read it.


Free (local build): the full pipeline on your machine, record, transcribe, label speakers, write notes, summarise, chat with your meetings. The local MCP connector is free too: point your own Claude or Codex at it and they read the meetings you choose, on request. Open source under AGPL-3.0. No account required.

Connected tier (AUD$20/yr): everything in the free build, plus:

  • End-to-end encrypted sync across your own devices.
  • The Android phone companion: capture audio and take notes on the go (iOS coming soon). Processing stays on your desktop; the audio and transcripts move over the encrypted sync channel.

Annual billing keeps the price lower than equivalent monthly billing would. We intend to raise the price after the launch period; subscribers at the current price keep it.


No. The free build makes no outbound connections other than the model download you initiate. The connected tier connects to the relay for sync, but we do not collect usage metrics, crash reports, or analytics from either build.


The live assistant is off by default, with three settings. Auto enables it only when a discrete GPU is present and GPU acceleration is active (on an integrated GPU it stays off). On enables it on any machine, accepting slower or heavier memory use. Off never runs it.


Does retrieval send my attachments anywhere?

Section titled “Does retrieval send my attachments anywhere?”

No. Attachments and the transcript are indexed on your machine. When you ask a question, the app retrieves the relevant passages locally and passes them to the local language model; nothing is uploaded. The text-embedding model that powers retrieval (~600 MB) is an on-demand download you start yourself, the same class as the existing model downloads. It does not change the free build’s network claim.


Minutist ships with local models: a speech recognition model for transcription, a language model for summarisation and chat, a vision model that reads text in image attachments, and a text-embedding model that powers on-demand retrieval. All run on your machine. The exact model files and versions are shown inside the app; you download them on first run, and updates are offered in-app when available.

We do not list specific model names here because they change between releases.


On your local filesystem, under your platform’s standard application data directory. The exact path is shown in Settings → Storage. Nothing is mirrored to a remote location in the free build.


Open a thread on the GitHub Discussions page.