Open-source meeting intelligence
Never take meeting notes again.
NoteAI sends a bot to your Google Meet calls, transcribes every speaker, and turns the conversation into summaries, action items, and decisions — automatically.
Quick recap — we agreed the beta opens to the first hundred teams next Thursday.
Right. I'll have the onboarding emails drafted by Monday so we can review them together.
Perfect. One open question: do we gate the export feature for beta users?
Let's keep exports on. It's the thing early users asked for the most.
How it works
From calendar invite to finished notes
You join the meeting. NoteAI does everything else.
- Step 1
Connect your calendar
NoteAI syncs upcoming Google Meet, Zoom, and Teams meetings via OAuth — no manual scheduling.
- Step 2
The bot joins for you
A silent bot joins as a participant and records audio. No extension, no app install for attendees.
- Step 3
Every speaker, transcribed
Your chosen STT engine (Whisper, Deepgram, Groq, or Aikonsh Indic) transcribes the call. pyannote diarization labels who said what.
- Step 4
Get the intelligence
Llama 3 (via Groq) writes the summary, action items, key questions, and chapters — ready before you close the meeting tab.
Platform support
Sits quietly in every meeting
NoteAI works across the platforms your team already uses — with more on the way.
Google Meet
Full support — bot joins, records, and transcribes automatically.
Zoom
Full support — the bot joins Zoom meetings and handles multi-speaker calls.
Microsoft Teams
In progress — core transcription works, full bot automation coming soon.
No browser extension required. The bot joins as a participant — nothing installs on attendees' machines.
Features
Everything a meeting leaves behind, kept
Transcripts, decisions, questions, and follow-ups — captured without anyone lifting a pen.
Auto-join bot
Joins Google Meet, Zoom, and Teams on schedule. Zero clicks during the call — no extension for attendees.
Speaker-aware transcripts
pyannote diarization labels each voice. Works with Whisper, Deepgram, Groq, and Aikonsh Indic — swap engines with one env var.
AI summaries & action items
Llama 3 (via Groq) distills the call into what matters — summary, action items, key questions, and chapter markers.
Chapters & key questions
Jump to the exact moment a decision was made or a question was raised.
Searchable history
Every meeting, filterable and full-text searchable across all platforms.
Export anywhere
Download transcripts as PDF, or export CSV/JSON for your own pipelines.
Manual upload
Drop a .wav or .mp3 to transcribe any recording without the bot.
Speech-to-text engines
Pick the transcription engine that fits
NoteAI is engine-agnostic. Swap between fully local models and cloud APIs from a single config — no code changes.
Whisper
Fully local, fully private
OpenAI Whisper runs entirely on your own hardware. No audio ever leaves your machine — ideal for confidential calls, legal, or healthcare meetings.
Deepgram
Real-time API, enterprise accuracy
Deepgram's Nova model delivers near-instant transcription via API with excellent accuracy across accents and noisy environments.
Groq (Whisper)
Whisper at GPU-cluster speed
Run Whisper-large through Groq's LPU inference — same model, dramatically faster. Ideal when you want local-quality transcription without waiting.
Aikonsh Indic
Hindi & Indian language support
An in-house model trained on Hindi and other Indic languages. Handles code-switching (Hinglish), regional accents, and domain-specific vocabulary out of the box.
RecallAI
Managed bot — zero config
RecallAI handles the bot joining, audio capture, and transcription as a single managed API. The easiest path to production — just add your key and go.
# choose one:STT_ENGINE=whisper # local, freeSTT_ENGINE=deepgram # fast APISTT_ENGINE=groq # fastestSTT_ENGINE=aikonsh # Hindi / IndicSTT_ENGINE=recallai # managed, easiestA closer look
Built around the way meetings actually work
Live transcript
Watch the conversation become text, speaker by speaker
While the bot sits quietly in the call, Whisper transcribes the audio and pyannote attributes every sentence to the person who said it. No more “who agreed to that?” — the transcript remembers.
Quick recap — we agreed the beta opens to the first hundred teams next Thursday.
Right. I'll have the onboarding emails drafted by Monday so we can review them together.
Perfect. One open question: do we gate the export feature for beta users?
Let's keep exports on. It's the thing early users asked for the most.
AI summary
The whole meeting, distilled into tabs you can act on
Minutes after the call ends, NoteAI writes the summary, lists every action item with an owner, and surfaces the key questions left open — so the follow-up writes itself.
Meeting history
Every meeting you've ever had, one search away
The dashboard keeps your full history — filter by date or attendee, or full-text search across every transcript to find the exact moment a decision was made.
Self-host
Yours to run. Free and open-source.
NoteAI is fully self-hostable. Clone the repo, pick your meeting platform and STT engine, and run it on your own machine — your audio and transcripts never leave your control.
- Django
- Playwright
- Whisper
- Deepgram
- Groq
- Aikonsh Indic
- pyannote
- Llama 3
- React
$ git clone https://github.com/ft-prince/NoteAI$ cd NoteAI# configure platform + STT engine in .env# PLATFORM=google_meet | zoom | teams# STT_ENGINE=whisper | deepgram | groq | aikonsh | recallai$ docker compose upFAQ
Questions, answered
Yes. When you use Whisper or Aikonsh Indic, all transcription happens on your own machine — no audio or transcript leaves your infrastructure. If you choose Deepgram or Groq, audio is sent to their APIs under your own key. RecallAI is a managed service with its own data policy.
Google Meet and Zoom are fully supported today — the bot joins automatically, records the call, and delivers transcripts and summaries. Microsoft Teams support is in active development and working in an early state. No browser extension is needed for any attendee.
NoteAI supports five engines, switchable via a single env var: Whisper (fully local, free), Deepgram (real-time API), Groq Whisper (fastest inference), Aikonsh Indic (in-house Hindi and Indian language model), and RecallAI (fully managed, easiest setup — paid). You can swap between them without changing any code.
Yes. Set STT_ENGINE=aikonsh in your .env to use the Aikonsh Indic in-house model, which is trained on Hindi and other Indic languages including Hinglish (Hindi/English code-switching). It handles regional accents and domain-specific vocabulary that generic models often miss.
Llama 3 served via Groq's LPU inference for summaries, action items, key questions, and chapters. Groq's ultra-fast inference means your meeting notes are ready seconds after the call ends rather than minutes.
Yes. NoteAI is MIT-licensed and open-source. With Whisper for STT and your own Groq API key (generous free tier), you can run the full stack at zero cost. RecallAI is the only paid option — and only if you choose it for convenience.