Live transcription, question detection and grounded answers, in an overlay only you can see. No bot joins the call and nothing shows up on a screen share.
★★★★★ 4.9 average · macOS and Windows · free tier, no card
Everything you need to stay sharp, confident and ahead on every call.
A native always-on-top window built with PyQt6, created with the operating system's capture-exclusion flag set. It is on your screen and out of every recording, screen share and screenshot. No browser tab, no extension, no second participant in the attendee list.
Rollback is a single flag, reversible for 72 hours. The old schema stays warm until sign-off.
Cmd Shift H ⌘ ⇧ H hide⌘ ⇧ ← shrinkmdash; hideCmd Shift L ⌘ ⇧ H hide⌘ ⇧ ← shrinkmdash; shrinkAmazon Transcribe streaming delivers partial results in roughly 280 milliseconds, with speaker labels, punctuation and a custom vocabulary you control. Product names, people and acronyms come back spelled the way your company spells them, not the way they sound.
Claude drafts a concise, professional answer from your own runbooks, proposals and past calls. It answers from your material or it says it does not know. Every claim carries the source it came from, so you can glance at the citation before you say it out loud.
FastAPI with async workers behind a WebSocket, so transcription, retrieval and generation never block one another. If a model call is slow the transcript keeps flowing. If the network drops, the buffer holds and the session stitches itself back together.
Start free, upgrade when you need it, cancel any time.
Starter
For trying it on your own calls.
Free forever. No card.
Free forever. No card.
Get started freePro
For people who are on calls all day.
Billed monthly. Cancel any time.
$180 billed once a year.
Start 14-day trialEnterprise
For teams with a security review.
Priced on seats and deployment.
Contact salesNo card for the free tier Cancel in one click Audio is never stored macOS and Windows
Five steps from your microphone to an answer on screen.
The desktop app captures system audio and your microphone through a loopback device. No bot joins the call and no one sees a second participant.
Audio streams to Amazon Transcribe over a persistent WebSocket in 100ms chunks, with speaker labels and punctuation coming back live.
A lightweight classifier watches the transcript for direct questions, objections and commitments, and tags them the moment they land.
Claude drafts a grounded answer from your own notes, docs and CRM records. Every claim cites the source it came from.
The answer appears in an always-on-top window excluded from screen capture. You see it. The call does not.
From your microphone to a grounded answer on screen, and the record it leaves behind.
Built on services your security team has already reviewed.
Native always-on-top overlay with capture exclusion on Windows and macOS.
Async workers, retries and backpressure so a slow call never blocks the stream.
Sub-second partial results with speaker diarisation and custom vocabulary.
Grounded answers from your own material, with guardrails and citations.
Session state in Redis, searchable transcripts and recaps in Postgres.
★★★★★
"I stopped taking notes in discovery calls. The recap it writes is better than mine and it is waiting for me when I hang up."
★★★★★
"The question detection is the part people do not expect. It catches the thing a client slipped in at minute 40 that I would have missed."
★★★★★
"Our screen is shared for most of every call. The overlay has never once turned up in a recording. That was the whole requirement."
Yes. The overlay is created as a capture-excluded window using the operating system's own APIs, the same mechanism password managers use to keep their windows out of screenshots. It is excluded from Zoom, Teams, Meet and Webex screen sharing and from OS-level recording. It is not invisible to somebody sitting beside you looking at your monitor, and it will appear if you point a physical camera at your screen.
You need whatever your jurisdiction and your company policy require, and that is on you rather than on us. Several regions are all-party consent. The app can announce itself at the start of a call, and an admin can make that announcement mandatory for everyone on the plan.
Partial transcription lands in about 280ms. A drafted answer typically appears 1.5 to 3 seconds after the question finishes, depending on how much of your own material it has to search. The transcript shows immediately either way, so nothing waits on the model.
Audio is streamed, not kept. It goes to the transcription service and is discarded once the text comes back. Transcripts and recaps are stored so you can search them, and on Enterprise you set the retention window, including zero. Nothing is used to train a shared model.
All of them. The app captures system audio rather than joining the call as a participant, so it works with Zoom, Teams, Meet, Webex, a phone on speaker, or a recording you play back. There is nothing to install in the meeting platform and no bot in the attendee list.
The audio buffer keeps recording locally and the backend holds the session open. When the connection returns, the buffered audio is sent and the transcript stitches back together. You lose the live answers for that window, not the record of what was said.
Install it, join your next call, and see the first answer appear before you have finished reading the question.
macOS 13+ and Windows 11 · 10 free minutes a month · cancel in one click
.png){kind=small}