Minutist

Where your data lives

Three channels. Three answers.

Minutist has three distinct data channels with different properties. This page states each one plainly. No single claim covers all three. Read the channel that applies to you.

Free build

Everything on your machine.

The free build is fully local. No account, no telemetry, no server. The entire pipeline (record, transcribe, label speakers, take notes, summarise, chat, live assistant, and transcript-aware retrieval) runs on your device. Source is published under AGPL-3.0.

The only network operation it makes is the model-weights download you start yourself. You pick the model, you start the download; nothing else leaves the machine. This includes the text-embedding model that powers retrieval; it is an on-demand download of the same kind, not a server call.

  • No account required
  • No telemetry of any kind
  • The MCP connector is a local endpoint on your machine; nothing reads it unless you point your own assistant at it
  • Model download is user-initiated and one-time per model
Connected · Sync

Ciphertext the server never reads.

The sync channel is end-to-end encrypted. Your device encrypts the data before it leaves; the relay server brokers the connection and store-and-forwards ciphertext so your other devices can pick it up when they come online.

The server never holds the plaintext, and the key never leaves your devices. This channel may accurately be called encrypted.

  • Encrypted on-device before transmission
  • Server stores and forwards ciphertext only
  • Key material stays on your devices
  • Available in the connected tier (AUD$20/yr)
Free · Connector

Straight from your desktop to your AI service.

The connector is a local endpoint on your own machine. Point your own Claude or Codex client at it and turn it on, and that client can read your meetings (transcripts, notes, and summaries) and send whatever it reads to the external AI provider you connected, so the assistant can answer. It is a direct path from the desktop app to your AI service; it never passes through a Minutist server.

While the connector is on, the assistant can list and read any of your meetings through it (not just one), which is what lets you ask about any of them. The AI provider receives and processes whatever the assistant reads, under its own policy. This channel is not encrypted end-to-end, and it is separate from sync.

It is free and local: no hosted relay sits in the path, and no account is required. The connector stays off until you enable it and point a client at it.

  • A direct path: the desktop app to your AI provider, never via a Minutist server
  • Off until you enable it and point your own client at it
  • While on, the assistant can read any of your meetings through it
  • The AI provider receives and processes what the assistant reads, under its policy
  • Not encrypted end-to-end; separate from the sync channel

At a glance

Channel summary

Channel What leaves your machine Who can read it When
Free build: model download The model weights you download, initiated by you You and your network path Once, on first use of each model
Connector (free build) The meeting content the assistant reads, full text You, and the external AI provider you connected While the connector is on, on the assistant's request
Sync (connected tier) Ciphertext only, encrypted before it leaves the device Only your devices (the server never holds the key) On any change, while you have a connected-tier subscription

Start with the free build: your machine, your data.