What is minutes?

minutes is a self-hosted tool for live transcription and translation. Join a Google Meet or Microsoft Teams call — or just open a webinar, a recorded video, or a podcast — and minutes turns the audio into a written transcript as it plays, and (if you want) translates each line into another language in real time. You can also record your own microphone right in the app (on a computer or a phone, no extension) or upload a recording you already have. The whole transcript streams into a clean web app you can read, search, rename, copy, share, and export.

It's built for teams who want this capability on infrastructure they run, not handed to a third-party meeting-bot service. minutes is multi-user, EU-friendly, and designed with privacy in mind.

Who is this for?

Two audiences. Operators / admins who deploy and run minutes for their organization, and end users who join meetings and read the live transcripts. The docs are split the same way — jump to Where to go next below.

The three parts

minutes is made of three pieces that work together:

1. The web app — the product UI. This is where you see live transcripts, turn translation on, rename meetings, copy, export, and create share links. It can also record your own microphone directly (no extension), and it installs to your phone's home screen as a full-screen app. It runs at / (and also /app) on your server.
2. The Chrome capture extension — a small browser extension (loaded from capture/extension) that captures the audio of any browser tab and streams it to your backend. Meet and Teams are auto-detected (and named with the real meeting ID); any other tab playing audio works too. You sign in through the extension's popup and start capturing the tab you're on.
3. The backend — a FastAPI service that runs the pipeline and stores everything. It ships as a single-box Docker Compose stack: a Caddy TLS edge, the backend itself, Postgres (durable transcripts and translations), Valkey/Redis (live fan-out and admission control), a MinIO object store (archived audio), and a scheduler for background work like upload transcription and retention purges.

Three ways to capture

However you capture, you get the same kind of meeting — the same live transcript, translation, copy, export, and sharing:

1. A browser tab's audio — the Chrome extension streams the audio of any tab (a Meet/Teams call, a webinar, a video) from your desktop browser. It can also add your own microphone as a separate track at the same time.
2. Your microphone, in the app — record directly in the web app, on a computer or a phone, with no extension and no second device. It's a mic-only capture (the Host mic source).
3. An audio file — upload a recording you already have and minutes transcribes it in the background.

How it works: the pipeline

The path from a spoken sentence to a translated line on your screen is short and only ever touches two outside services:

Audio in  (a browser tab, your microphone, or an uploaded file)
        │  (the extension, in-app recording, or an upload streams it in)
        ▼
   minutes backend  ──►  Soniox  (speech-to-text: en / de / fa)
        │                         returns the transcribed text
        ▼
   minutes backend  ──►  Anthropic / Claude  (optional translation)
        │                                      returns the translated text
        ▼
   Live transcript in the web app  (searchable, exportable, shareable)

1. Capture — audio reaches the backend over a secure WebSocket, whether from the extension (a tab), in-app recording (your mic), or an upload.
2. Speech-to-text — the backend sends audio to Soniox, which returns the words.
3. Translate — each finalized line is optionally sent to Anthropic (Claude) and translated into the meeting's chosen output language.
4. View — the transcript (and translation) streams live into the web app.

Only two services ever see your data

Soniox receives audio (for transcription) and Anthropic receives text (for translation). Everything else stays on the box you operate — Postgres, Redis, object storage, and the app itself. For GDPR, host the box in an EU region and select Soniox's EU data-residency region so the audio is processed in the EU too (see Configuration → Data residency).

Supported languages

minutes recognizes and works with three languages:

Language	Code	Notes
English	en
German	de
Persian	fa	Right-to-left (RTL); rendered correctly throughout

Speech is transcribed in these languages, and translation can target one output language per meeting.

What you can do with it

• Live transcription of any tab you capture — a Meet/Teams call, a webinar, a video — as it happens.
• In-app recording — record your own microphone right in the web app, on a computer or a phone, with no extension and no second device. It's a mic-only Host mic capture that streams to the same pipeline and becomes a normal meeting. Great for dictation, a voice note, or an in-person talk.
• Dual-source capture — with the extension, record the tab's audio (Online stream) and your own microphone (Host mic) at once, kept fully separate (their own transcripts and timestamps). In Meet/Teams your own voice isn't played into the tab, so the Host mic adds your side — no second bot account.
• Translation — one output language per meeting, with an optional custom prompt and model, plus on-demand "translate this line" for a single line.
• Manage meetings — rename them and read absolute timestamps for every line.
• Copy the whole transcript (or translation) for the source you're viewing, as plain text with no timestamps — a one-click clipboard grab for pasting elsewhere.
• Export transcripts as txt, md, or json; choose transcript, translation, or both, with or without timestamps.
• Audio upload — transcribe an existing recording (audio or video container) instead of a live call. Processed in the background by the scheduler.
• Public share links — share a read-only view via an opaque token you can rotate or disable.
• Install on your phone — the web app is an installable PWA that runs full-screen from your home screen, and it's fully responsive, so you can record, read, translate, copy, share, and export from a phone. (The capture extension itself runs on desktop Chrome.)
• GDPR controls — an optional consent gate, a configurable retention purge, and an erasure path for owners and admins.

Accounts are admin-created

There is no public signup. An admin creates each user from the command line, and users set their own API keys in the web app's Settings. See Getting started for what a user needs, and Deploy for the admin side.

Privacy and self-hosting at a glance

• You run it. A single VPS with Docker runs the whole stack.
• Two external processors only: Soniox (audio) and Anthropic (text). No third-party meeting bot joins your calls — capture happens in your browser.
• EU-friendly. Built with GDPR in mind: consent gate, retention limits, and data erasure — plus EU data residency end to end when you pick Soniox's EU region.

Where to go next

Pick the path that fits you:

• Admins / operators → head to Deploy to stand up the single-box stack and create your first users.
• End users → head to Getting started to set up your keys and start your first live transcript — capture a tab with the extension, or just record your microphone in the app.