NoteCast — AI tab-audio note-taker

Capture a browser tab's audio, transcribe it locally with Whisper, and generate structured notes with a local Ollama model.

NoteCast turns anything you can hear — lectures, meetings, videos, podcasts, or an audio file on your disk — into a live transcript and structured notes, without sending a single second of audio anywhere. Your machine, your models Audio is transcribed on-device with Whisper (WebGPU, falling back to WASM). It never leaves your computer. Notes are written by Ollama — running locally by default (http://localhost:11434), or on a remote server you run yourself. No third-party cloud, no account, no vendor API keys, no telemetry. Heads up: notes need Ollama. Transcription works on its own with zero setup, but the notes half needs Ollama installed with a model pulled. NoteCast walks you through it on first run — it detects whether Ollama is reachable, pulls the model for you from inside the extension, and if Ollama needs a one-time config tweak to accept the extension it shows the exact command to copy. What you get - Live notes: the transcript grows about every 30 seconds and the notes regenerate as you watch — Summary, Key decisions, Key points, Action items, Open questions. Regenerate on demand, including for past sessions. - Transcribe a file: run an mp3, wav, m4a, ogg, flac, mp4, webm, or mov from your device through the same pipeline, with a progress bar and time remaining. Nothing is uploaded. - Timeline-aware: transcript lines are clickable timestamps. On YouTube they deep-link to the exact moment; elsewhere they seek the tab's own player. Navigating to a new video mid-session splits the session automatically. - Pause and resume mid-session; pop the panel out into its own window. - Pick your quality: Base, Small, Medium, or Large (v3-turbo), multilingual or English-only, with the download size shown up front — and remove downloaded models from Settings when you want the disk space back. - Speaks your language: 28 interface languages (with right-to-left layout), 28 transcription languages with auto-detect, and notes written in whichever language you choose. - Session history, keyboard shortcut, Copy as Markdown, and Markdown or plain-text export with action items as checkboxes. Requires Chrome, Brave, or Edge 116+. Ollama with a model pulled (e.g. ollama pull gemma3:4b) — local by default, or a remote host you set in Settings. First use of each quality downloads the Whisper weights from Hugging Face once, then caches them. On Edge, assign the Ctrl+Shift+E shortcut yourself at edge://extensions/shortcuts — Edge does not bind it automatically. Privacy Your audio is transcribed on-device and never leaves your machine. Transcripts, notes, and settings are stored locally. If you point NoteCast at a remote Ollama host, only the transcript text goes to that server — one you run yourself. No analytics, no telemetry, nothing sent to the developer. See the privacy policy for details.