If you’re looking for a meeting recording API, consider checking out Recall.ai, an API that records Zoom, Google Meet, Microsoft Teams, in-person meetings, and more.
Sponsored by Recall.ai — The unified API for meeting bots.
Upload a meeting recording, get structured minutes back. Transcription and summarisation run on providers you choose — a hosted API, or a model running on your own machine. Everything is configured from the app itself, and everything it stores stays on your computer.
Runs anywhere Docker runs: Windows, macOS, Linux.
- Your choice of transcription — Deepgram, AssemblyAI, ElevenLabs Scribe, Sarvam AI, any OpenAI-compatible endpoint (OpenAI, Groq, Speaches, LocalAI), or Whisper running locally with no API key
- Your choice of LLM — any OpenAI-compatible API: OpenAI, OpenRouter, Groq, Together, Fireworks, Mistral, Cerebras, Anthropic, Gemini, or Ollama / LM Studio / vLLM on your own hardware
- Speaker-labelled transcripts so decisions and action items get attributed correctly
- Any meeting length — long recordings are transcribed and summarised in parts, then merged, so neither the upload limit nor the model context window caps you
- Real background jobs — Celery and Redis, with live progress, an activity log, cancel and retry
- Configured from the screen — API keys, models, prompts, retention: no editing files, no restarts
- Nothing phones home — no CDNs, no fonts, no telemetry. The only outbound calls are to the AI provider you configured
- Built to run — health and readiness endpoints, Prometheus metrics, structured logs, automatic retention cleanup and stale-job recovery
git clone https://github.com/inboxpraveen/LLM-Minutes-of-Meeting.git
cd LLM-Minutes-of-Meeting
docker compose up --buildOpen http://localhost:5000. That is the whole setup — no .env file, no API keys on disk.
The first screen asks you to create an administrator account. After that, go to Settings and add an LLM endpoint and a transcription provider. Press Test on each to confirm they work before you upload anything.
Want transcription with no API key at all? Build with Whisper included:
WITH_LOCAL_WHISPER=true docker compose up --buildThen choose Local Whisper as the speech provider. On Windows PowerShell, use
$env:WITH_LOCAL_WHISPER="true" on the line before.
You need Python 3.10 or newer, ffmpeg on your PATH, and Redis.
pip install -r requirements.txt
redis-server # terminal 1
celery -A celery_worker:celery worker -l info # terminal 2
celery -A celery_worker:celery beat -l info # terminal 3
python run.py # terminal 4Optional: pip install -r requirements-local.txt adds local Whisper.
Both choices are made in Settings. Neither is locked in — switch whenever you like.
| Provider | Speaker labels | Notes |
|---|---|---|
| Deepgram | Yes | Fast and accurate. Default model nova-3. |
| AssemblyAI | Yes | Strong accuracy, straightforward pricing. Default model universal. |
| ElevenLabs Scribe | Yes | Very good on multilingual audio. Handles stereo recordings per channel. |
| Sarvam AI | Yes | Best for Indian languages and code-mixed speech. Uses the batch API, so length is unlimited. |
| OpenAI-compatible | No | Anything exposing /v1/audio/transcriptions: OpenAI, Groq, or a local server. Long files are split automatically. |
| Local Whisper | No | Runs on your machine. No API key, no per-minute cost, nothing leaves the computer. |
Any OpenAI-compatible chat API. The Settings screen has presets that fill in the base URL for you:
| Provider | Base URL |
|---|---|
| OpenAI | https://api.openai.com/v1 |
| OpenRouter | https://openrouter.ai/api/v1 |
| Groq | https://api.groq.com/openai/v1 |
| Together AI | https://api.together.xyz/v1 |
| Fireworks AI | https://api.fireworks.ai/inference/v1 |
| Mistral | https://api.mistral.ai/v1 |
| Cerebras | https://api.cerebras.ai/v1 |
| Anthropic | https://api.anthropic.com/v1/ |
| Google Gemini | https://generativelanguage.googleapis.com/v1beta/openai/ |
| Ollama | http://localhost:11434/v1 |
| LM Studio | http://localhost:1234/v1 |
Running Ollama or LM Studio on your machine while the app is in Docker? Use
host.docker.internalinstead oflocalhost.
Use your API key with the matching base URL. An OpenRouter key on OpenAI's URL returns a 401.
Everything the app stores — the database, uploaded recordings, logs, and the key that encrypts your API keys — sits in one directory on your machine:
| Platform | Location |
|---|---|
| Windows | %LOCALAPPDATA%\MoM-AI |
| macOS | ~/Library/Application Support/MoM-AI |
| Linux | ~/.local/share/MoM-AI |
| Docker | the app_data volume, mounted at /data |
Back up that directory and you have backed up the whole application. Set MOM_DATA_DIR to move it.
API keys are encrypted at rest with a key generated on first run and stored in the same directory.
| Endpoint | Purpose |
|---|---|
/healthz |
Liveness. Returns 200 whenever the web process is up. |
/readyz |
Readiness. Checks the database, the job queue and ffmpeg. |
/metrics |
Prometheus metrics: request rates, job durations, meetings by status. |
Admin → System shows the same checks in the browser, plus disk usage, job counts, recent failures and the tail of the application log.
Optional Celery dashboard:
docker compose --profile monitoring upFull details are in docs/operations.md.
| Guide | Contents |
|---|---|
| docs/configuration.md | Every setting, and the handful of environment variables |
| docs/providers.md | Getting a key for each provider, and running models locally |
| docs/operations.md | Monitoring, backups, upgrades, scaling, reverse proxies |
| docs/troubleshooting.md | What to do when something does not work |
app/
├── __init__.py App factory, error handlers, Celery wiring
├── config.py Process-level configuration
├── settings.py Runtime settings: the schema behind the Settings screen
├── security.py Local key file, encryption of stored secrets
├── paths.py OS-specific data directory
├── models.py User, Meeting, MeetingEvent, SystemConfig
├── health.py Dependency checks for /readyz and Admin → System
├── logging_setup.py Structured logging with request ids
├── metrics.py Prometheus collectors
├── auth/ First-run setup, sign in, account
├── dashboard/ Home
├── meetings/ Upload, list, detail, cancel, retry, download
├── admin/ Settings, users, system
├── api/ Status polling, health, metrics
├── tasks/ process (the pipeline) and maintenance (retention, stale jobs)
├── providers/
│ ├── llm/ OpenAI-compatible client and chunked summarisation
│ └── speech/ Deepgram, AssemblyAI, ElevenLabs, Sarvam, OpenAI-compatible, local Whisper
├── utils/ ffmpeg helpers, markdown rendering and sanitising
├── static/ Compiled CSS and JS, no CDN
└── templates/ Jinja2 templates
tests/ pytest suite
pip install -r requirements-dev.txt
pytestThe stylesheet at app/static/css/tailwind.css is committed so the app needs no build step. If you
change templates, rebuild it:
npm install && npm run build:css
# or, with no Node installed:
docker run --rm -v "$PWD":/work -w /work node:20-alpine \
npx -y tailwindcss@3.4.17 -c tailwind.config.js \
-i app/static/css/tailwind.src.css -o app/static/css/tailwind.css --minify