Skip to main content

Getting started

minutes is a self-hosted, EU-friendly tool for live transcription and translation. It captures the audio of any browser tab — a Google Meet or Microsoft Teams call, a webinar, a video — transcribes it as it plays (English, German, Persian), and — if you give it an Anthropic key — translates each line into a language you choose. Everything runs on infrastructure your organization controls.

There are three pieces:

  • A web app — the product UI where transcripts appear live, get translated, copied, exported, and shared. It can also record your own microphone directly (no extension) and install to your phone's home screen as a full-screen app.
  • A Chrome capture extension — streams the audio of your Meet/Teams browser tab to the backend.
  • A FastAPI backend — the single-box Docker stack (Caddy TLS edge, backend, Postgres, Valkey/Redis, MinIO object store, and a scheduler) that runs the pipeline and stores everything.

The pipeline in one line: the extension captures tab audio → Soniox turns it into text → Anthropic / Claude optionally translates it → the live transcript shows up in the web app. Only those two external services ever receive your data; everything else stays on the operator's box.

Which track are you?

This page splits into two short tracks. Pick the one that matches you — you do not need to read both.

  • Administrators stand up the server and create accounts. → Track A
  • Users sign in, set their keys, and capture meetings — with the extension or by recording in the app. → Track B

Who needs what​

A quick map of who touches which part:

  Administrator                              User
  ─────────────                              ────
  • A domain + an EU VPS                      • An account (the admin makes it)
  • Docker + the minutes repo                 • Their OWN Soniox key + region (live + uploads)
  • Creates user accounts via CLI             • Their OWN Anthropic key (for translation)
  • (optional) a fallback Soniox key          • A way to capture: the extension (a tab),
                                                in-app recording (their mic), or an upload
        │                                              │
        └──────────────►  one minutes server  ◄────────┘
                          (web app + backend)

The single shared resource is the server. The admin runs it (and may set an optional fallback Soniox key); each user brings their own keys — Soniox for transcription, Anthropic for translation — for the features that bill to them.

Track A: Administrators​

You are technical and you run the box. Three things get you from zero to a working server.

1. Deploy the server​

minutes ships as a one-VPS Docker Compose stack. The short version:

git clone <your-fork-of-minutes> && cd minutes/deploy/single-box
cp .env.example .env
# fill in .env (DOMAIN, MINUTES_SONIOX_API_KEY, passwords, the two secrets)
docker compose up -d --build

A one-shot migrate service runs alembic upgrade head before the backend starts. To update later: git pull && docker compose up -d --build. Caddy auto-provisions a Let's Encrypt TLS certificate for your $DOMAIN, so inbound TCP 80 and 443 must be public.

A real domain is required

The capture extension opens a secure WebSocket (wss://) from the browser's secure context. That needs TLS, which (via Caddy + Let's Encrypt) needs a real public domain with an A record pointing at the VPS. A bare IP or http://localhost only works for local development (ws://, no TLS) — the extension and the HTTPS web app will not work without a domain.

The full walkthrough — sizing, prerequisites, boot order, verifying, operating, and backups — is on the deploy page.

→ Deploy a server

2. Create user accounts​

There is no public signup. You create every account from the CLI inside the backend container:

docker compose exec backend python -m app.admin create-user --email you@org.com [--admin]

The first user you create is automatically an admin. Other commands: set-password --email, delete-user --email, and list-users.

→ Manage users

3. Hand out the basics​

Tell each new user their server URL and that you've created their account, then point them at Track B. They will set their own password, plug in their own keys, and install the extension.

What the server pays for vs. what users pay for

Both keys are per-user: each user sets their own Soniox key (transcribes their live captures and uploads) and their own Anthropic key (translation) in the web app. The optional MINUTES_SONIOX_API_KEY in .env is only a fallback for live capture when a user hasn't added their own — handy for a quick demo, but leave it empty to require per-user keys. See Configuration → keys.

Track B: Users​

You are here to capture meetings. No server setup — your administrator has already done that and given you a server URL and an account.

1. Sign in​

Open the web app your admin gave you (typically https://your-domain/app) and sign in with your email and password. This is also where everything you capture will appear, live.

→ Using the web app

2. Set your keys​

In the web app's Settings, add your own API keys. These unlock the features that bill to you:

  • Soniox key (soniox.com) — powers transcription, both your live captures and audio files you upload. Also pick your Soniox region (US/EU) next to the key — that's where your audio is processed.
  • Anthropic key (console.anthropic.com) — needed for translation (live, on uploads, and the on-demand "translate this line"). No key simply means translation is off.
note

Transcription uses your Soniox key. If your admin set a shared fallback key, live capture still works before you add your own — but uploads always need yours, and your own key lets you choose your data region.

3. Capture something​

There are two ways to capture live audio (plus uploads). They produce the same kind of meeting — pick whichever fits.

Option A — Record your own voice, right in the app (the quickest start)​

No extension, no second device. In the web app, click Record (in the Transcriptions list header on a computer, or the prominent Record action on a phone), grant the microphone, run a quick level test, then press Record. A live view shows the timer, level meter, Stop, and the transcript as you speak. Press Stop and it becomes a normal meeting. This is a mic-only capture — your own voice (the Host mic) — so it's perfect for dictation, a voice note, or an in-person talk.

It works on a computer or a phone over your HTTPS site, and you can install minutes to your home screen so it runs full-screen like a native app.

→ Record in the app

Option B — Capture a browser tab with the extension (for calls and videos)​

To transcribe a Meet/Teams call, a webinar, or a video — i.e. a browser tab's audio — use the Chrome extension:

  1. Load the Chrome capture extension, open its settings (the gear icon in the popup), and enter your Server URL. The popup signs you in with your email and password, then mints a short-lived token for each tab it captures.
  2. Join your Google Meet or Microsoft Teams meeting as you normally would, in Chrome, with your own account. (minutes captures that tab's audio — you have to actually be in the call.)
  3. Open the extension popup on the meeting tab and start capture.
  4. (Optional) Toggle Also capture my microphone to add your own voice as a second, separate track (Host mic) — in Meet/Teams your own voice isn't played back into the tab, so this is how you get your side transcribed. No second bot account joins the call.

→ Install the extension

Either way, watch the live transcript — and, if you set an Anthropic key, the translation — appear in the web app. From there you can translate individual lines on demand, rename the meeting, copy the text (no timestamps), export it (txt / md / json, transcript / translation / both, with or without timestamps), and create read-only public share links.

→ Using the web app for the full feature tour.

tip

Have a recording instead of a live call? Upload the audio or video file in the web app — minutes transcribes it out-of-band using your Soniox key. Uploads are capped at roughly 300 MB each, with at most two processing at once.