Timestamped
Every line has a timecode. Chips jump back to the moment.
v0.1.7
Android APK just shipped
Every video becomes a searchable transcript.
Whisper runs on your machine. Paste a link, get a timestamped transcript and a clean summary.
YouTubeBilibiliVimeoTikTokXSoundCloudTwitchInstagramDouyinFacebooklocal files
$0
per minute
12h
longest recording
6
summary styles
0
accounts, ever
Tour
A whole app in one window.
408 segments. Search any word, copy any line.
See the model, the engine, and whether CUDA ran.
One field. Two clicks.
The extension takes five steps.
Features
Small app. Sharp edges.
Six summary styles
Brief, Standard, Detailed, Study Notes, Key Points, Meeting notes.
GPU, or not
CUDA when you have it. CPU when you don't.
Exports that fit
TXT, SRT, WebVTT and JSON.
One click from Chrome
Send the video you're watching with Alt+Shift+S.
Agents over MCP
A stdio bridge for Claude, Cursor or your own agent.
Download
Pick your shape.
Android, standalone
No account, no pairing, no desktop nearby. Long recordings are split and merged back.
Android 8.0+ · signed release.
Chrome extension
Reads the page you're on and sends the video to the app. Alt+Shift+S works anywhere.
chrome://extensions → Developer mode → Load unpacked.
youtube.com/watch?v=…
MCP server
The agent gets a transcript and summarises with its own model. No Gemini key involved.
stdio · local Whisper · 12 tools
docker build -t sonicbrief-mcp:local .
docker run --rm -v sonicbrief-data:/data \
sonicbrief-mcp:local --self-testOn the phone
Same app. Smaller screen.
Privacy
Local, and honest about it.
Bound to localhost
Nothing is reachable from another machine.
Key stays local
Encrypted with the Android Keystore.
Temp audio deleted
Only the transcript stays.
Nothing phones home
No analytics, no ads, no cookies.
Local Whisper keeps audio on the machine. Cloud summaries are opt-in, with your own key — that is the only step where anything leaves.
FAQ
Fair questions.
Everything else is in the README.
Do I need a GPU?
No. CPU works out of the box, just slower. An RTX card with the CUDA pack is several times faster.
Do I need a Gemini key?
Only for cloud summaries and the Android app. Local transcription never needs one, including over MCP.
How long can a recording be?
Up to 12 hours. The Android app splits long audio into twenty-minute ranges and merges the result.
Where are my transcripts?
In SQLite on your machine, or in the app's private storage on Android. Nothing is uploaded to us — there is no us.
Can my agent use it?
Yes, over stdio MCP. The agent gets the transcript and summarises with its own model.
Is it really free?
Yes. MIT for the app, GPL-3.0-only for android/. You supply a key only if you want cloud summaries.