Calls drift. CASE taps only you.
A private meeting copilot. It listens to both sides of a call on your machine, transcribes locally, and beeps with a small card nobody else on the call sees when the conversation starts going in circles.
Free and open source. Nothing leaves your machine unless you point it at a cloud model.
Four reasons it speaks up. Otherwise, silence.
Every twenty seconds CASE reads the last three minutes of the call and asks the model one question: is there a reason to speak up right now? Most of the time the answer is no, and you hear nothing. When it is yes, you get a beep and a card. Nobody else does.
-
Going in circles Third pass on Friday vs. the payments fix. Ask who owns the call.
The same point has been restated with no decision. CASE names it so someone can close it.
-
Someone is stuck Priya has gone quiet since the schema question. Offer to pair on the migration after the call.
A participant has stopped moving because they do not know the next step.
-
Quick fact The brief says the beta cutoff is 14 November, not the 21st. From Launch › brief.md.
A detail is missing and it is in your project sources. CASE looked it up.
-
An idea Last week’s session parked the feature-flag route. It fits what Sam just described.
An alternative nobody has raised, often from an earlier session of the same thread.
Transcript lines and names on this page are fictional stand-ins. The four kinds, the timings, and the card are the real product.
It never joins the call, and nothing leaves your machine.
Both sides, transcribed locally
Your microphone and the system audio of everyone else run through local speech models on your computer. Each line is tagged as you or them. There is no bot in the participant list.
Your model, your choice
Point the assistant at a local model through Ollama or LM Studio and nothing leaves the machine. Use a cloud provider with your own key if you prefer. One provider runs the observer, summaries, and vision.
Only you hear it
The beep plays on your side. The card floats above the call on your screen. CASE never speaks into the call. Whether to say the thing out loud stays your decision.
| Platform | Microphone | System audio |
|---|---|---|
| macOS | Working | Working on 14.2 and later. Asks for the audio capture permission once. |
| Windows | Working | Built, not yet verified on real machines. |
| Linux | Working | Built, not yet verified on real machines. |
A meeting is a thread, not a file.
End a meeting today and continue it tomorrow. Each session ends with a short summary that merges into the thread’s rolling summary, so the assistant walks into the next call already knowing what was decided.
Meetings live in projects. A project holds a brief and sources: pasted notes, local files and folders, checked-out repos, links, images. The assistant starts each check with only the brief and the last summary, and reads sources on its own when the conversation calls for it.
One shortcut runs the whole loop.
Press it once to start a meeting, once more to end it. The meeting starts unfiled; the assistant files it into a project once it knows what it is about. The tray menu, the command line, and a Unix signal do the same thing, so it fits whatever starts your calls.
Twenty seconds, start to card.
Put it beside your next call.
Windows and Linux builds are on the same releases page. System audio on those platforms is built but not yet verified.