Drop a recording, record live, or paste a transcript. A local AI writes the brief: summary, decisions, and action items. Your meeting never leaves this device.
Audio or video file: the Zoom download, a voice memo, an interview.
Capture the meeting live from your microphone, right in this tab.
Zoom or Teams already made one? Skip straight to the notes.
Every meeting produces two artifacts: a recording nobody replays and notes nobody finished writing. The tools that fix this, the cloud AI note-takers, work by sending your meeting to their servers, which is exactly where a confidential standup, a client call, or a performance conversation should never go. Many companies now prohibit them outright.
This tool does the same job without the upload. OpenAI's Whisper transcribes the recording in your browser, then a local language model (Qwen 2.5 or Phi 3.5, running on your graphics chip through WebLLM, or on your CPU when WebGPU is unavailable) reads the transcript and writes the brief: a TL;DR, the decisions made, the action items, and the topics covered. The models download once, cache in your browser, and work offline afterward. Nothing you say in a meeting ever reaches a server.
A meeting recording is among the most sensitive files you handle: it contains voices, names, candor, and sometimes things said about people who are not in the room. Cloud note-takers ask you to trade that for convenience. Running the models on your own device removes the trade: the convenience stays, the exposure goes. If you want proof, load the models once, turn off your wifi, and generate a brief anyway.
Drop the recording here and both steps happen on your own device: OpenAI's Whisper model transcribes the audio in your browser, then a local language model reads the transcript and writes a brief with a summary, the decisions made, action items, and topics. The recording, the transcript, and the notes never touch a server.
That is the reason this tool exists. Cloud AI note-takers process your meeting on their servers, which many employers and clients prohibit. Here the AI models download to your browser once and run locally, so a confidential meeting stays as private as it was in the room. You can even load the models once and then use the tool with your internet disconnected.
A short TL;DR summary, the decisions that were made, action items (as checkboxes you can tick off, with owners when they are clear from the conversation), and the topics discussed, followed by the full timestamped transcript. Everything is editable in place before you copy or download it, and a detail control lets you regenerate briefer or more thorough notes without re-transcribing.
Yes, and it is the fastest path: choose Paste a transcript and drop in the text or the .vtt/.srt file your meeting app produced. The tool skips transcription entirely and goes straight to writing the brief, which takes well under a minute on most machines.
A recent laptop handles it comfortably. With WebGPU (current Chrome, Edge, Firefox, or Safari) the language model runs on your graphics chip and writes the brief in seconds. Without WebGPU the tool automatically switches to a compact model on your CPU, which is slower but works everywhere and is just as private. The one-time model downloads are cached, so later visits work offline.
Yes. There is no account, no per-minute meter, and no monthly quota, because your own device does all the work. Cloud note-takers charge monthly precisely because they pay for the servers; this tool has no servers to pay for.