AI Meeting Summary Generator — Free, No Upload
Transcribe any meeting recording and get a TL;DR, key discussion points, and action items — powered by Whisper + SmolLM2 running entirely in your browser.
Drop your audio or video file here
MP3, WAV, M4A, MP4, WebM, MOV, OGG, FLAC · up to 150 minutes
How the AI Works
Whisper transcribes your audio
OpenAI's Whisper base model (75 MB) converts speech to text in 30-second chunks. It auto-detects the language and handles background noise, accents, and overlapping speakers better than older speech-to-text engines.
SmolLM2 generates the summary
HuggingFace's SmolLM2-1.7B-Instruct reads the full transcript and generates a structured summary: a TL;DR paragraph, 5–8 key discussion points, and a list of action items. The model is quantized to 4-bit (≈450 MB) and runs locally.
Everything stays on your device
Both models run as Web Workers in your browser tab via WebAssembly. Your audio file, transcript, and summary are processed entirely locally — no cloud API, no data transmission, no logs.
Models are cached after first download
Whisper and SmolLM2 are stored in your browser's IndexedDB cache. Subsequent sessions start immediately without re-downloading. If you already used the transcribe-audio tool, Whisper is already cached.
How to Summarize a Meeting Recording
- 1
Open the tool
No account or install needed — Whisper and SmolLM2 run entirely in your browser. Models are cached after the first download.
- 2
Drop your meeting recording
Drag and drop your MP3, WAV, M4A, MP4, or Zoom/Teams/Meet recording onto the upload area. Audio is decoded locally — nothing is uploaded.
- 3
Wait for transcription
Whisper AI transcribes the audio in 30-second chunks. Progress shows the current chunk and time position. A 10-minute meeting takes roughly 1–3 minutes on CPU, faster on WebGPU.
- 4
AI generates the summary
SmolLM2-1.7B-Instruct reads the transcript and produces a TL;DR paragraph, a list of key discussion points, and a checklist of action items — all locally in your browser.
- 5
Copy or download your notes
Copy individual sections (TL;DR, key points, or action items) or download the full summary as a .txt file. Toggle the transcript view if you need the verbatim text too.
What You Can Summarize
📹 Zoom / Google Meet / Teams
Export the recording as MP4 or MP3 and drop it in — the tool extracts the audio track automatically.
🎙️ Podcast interviews
Get the key talking points from a long-form interview or panel discussion in under a minute.
📚 Lecture recordings
Compress a 1-hour lecture to a page of notes. Review key concepts and any homework or reading assignments.
🤝 1:1 or team stand-ups
Turn a voice-memo or quick sync recording into a shareable summary your team can act on.
🎤 Conference talks
Summarize a keynote or breakout session recorded on your phone or downloaded from YouTube.
🗒️ Client calls
Generate meeting notes with action items you can paste into your CRM, email, or project tracker.
Frequently Asked Questions
How does the AI meeting summarizer work?
The tool runs two AI models locally in your browser. First, OpenAI's Whisper base model (75 MB) transcribes your audio to text using the same approach as the transcribe-audio tool. Then SmolLM2-1.7B-Instruct (450 MB, quantized to q4) reads the transcript and generates a structured summary — TL;DR, key points, and action items. Both models download once and are cached for future sessions.
Is my meeting recording private?
Yes — completely. Neither the audio file, the transcript, nor the summary is ever sent to any server. Whisper and SmolLM2 both run as WebAssembly workers inside your browser tab. The only outbound network traffic is the one-time model download from the HuggingFace CDN.
What audio and video formats are supported?
The tool accepts any file format your browser can decode — typically MP3, WAV, M4A, AAC, OGG, FLAC for audio, and MP4, WebM, MOV for video. It extracts the audio track automatically from video files, so Zoom recordings, Teams recordings, and screen captures all work without any conversion step.
How long can the meeting recording be?
The transcription step supports up to 150-minute recordings (the same limit as the dedicated transcribe-audio tool). The summarizer then truncates the transcript to roughly 6,000 tokens (~24,000 characters) before sending it to SmolLM2. For very long meetings, the summary is built from the first portion of the transcript.
How large is the model download?
Whisper base (q8) is 75 MB. SmolLM2-1.7B-Instruct (q4) is approximately 450 MB. Both download on first use and are cached in your browser's IndexedDB — subsequent runs start immediately without re-downloading. If you already used the transcribe-audio tool, the Whisper model is already cached.
Can I use this for Zoom, Google Meet, or Teams recordings?
Yes. Export your recording as an MP4 or MP3 file (Zoom saves locally; Meet and Teams send a download link), then drop it into the tool. The tool decodes the audio track, transcribes it with Whisper, and produces a structured meeting summary including action items.
How is this different from just using the transcribe-audio tool?
The transcribe-audio tool gives you the full word-for-word transcript with timestamps, which is ideal for subtitles, notes you want to search, or recordings where accuracy matters most. The audio-summary tool skips the timestamps and runs a second AI pass to compress a 60-minute transcript into a TL;DR paragraph, a bullet-point list of key topics, and a checklist of action items — saving you the time of reading the full transcript.