MimicScribe — In-Meeting Technical Assistant and Transcription for macOS

• ✦ Every speaker named — 97% accurate, no meeting bots required — search by speaker ^benchmark
• ✦ Talking points — What to ask next, action items refined as you talk — ~1 second
• ✦ Keyboard-first — Reachable from any app, no hunting through windows mid-meeting
• ✦ On-device speech processing — Audio never leaves your Mac

For people who end up in fast, packed meetings.

A system-level shortcut, like Spotlight or Raycast.

Assistant Prep Notes

Discovery call with Brightwave, a performance marketing agency. Goal: learn how they build client reporting today, what’s forcing a change and by when, and leave with a scoped next step. I build on their own stack — n8n, their warehouse, dashboards — not a tool they rent.

brightwave-account-brief.pdf client Looker Studio

Yes — free, within fair use. Recording, transcription, and summaries with named speakers are unlimited, with no account or card. The live in-meeting assistant — briefings and follow-up questions — has a free daily allowance that a subscription lifts. The limits exist to stop abuse, not to meter normal work — including a lifetime ceiling that's generous but not infinite.

Yes. Most AI notetakers join the call as a participant — a bot in the attendee list, a recording badge. MimicScribe captures system audio at the OS level instead, so it works with Zoom, Teams, Google Meet, Slack huddles, or any app that plays sound. Speakers are still separated on-device, just without anything joining the call. No plugins, no browser extensions.

Yes. The speech model recognizes 25 languages including English, Spanish, French, German, Italian, Portuguese, and more. Meetings that are primarily English with occasional code-switching — a customer quote in Spanish, a German product name, a French aside — transcribe cleanly and stay searchable. Technical vocabulary, proper nouns, and jargon work without a custom dictionary.

Transcription always works offline — it runs entirely on your Mac. AI features like text refinement, meeting summaries, and the meeting assistant require an internet connection. If you lose connectivity mid-meeting, transcription continues uninterrupted and AI enrichments are queued until you're back online.

• Microphone — required for all voice features.
• System Audio Recording — captures audio from video calls for meeting recording. Audio only — no screen capture or video.
• Accessibility optional — voice editing only. Skip it and meeting recording still works.

Recording laws vary by state and country — some US states and most of Europe require all parties to consent. MimicScribe doesn't announce itself to meeting participants; that's your call. Best practice: tell attendees at the top of the call, and stop recording if anyone objects. If you're in a regulated industry, check with your legal or compliance team before rolling it out.

No. MimicScribe holds at about 330 MB during a meeting — less than a single browser tab — and the heap doesn't grow as the recording runs long. ML models run on the Neural Engine, not main RAM, so your other apps aren't competing for memory. See full performance details →

Speech recognition, speaker separation, and echo cancellation run on your Mac. Speaker attribution, action items, summaries, and real-time suggestions require larger models than what can run locally with acceptable accuracy — so these are handled by cloud AI using transcript text only. Our proxy keeps no copies, and Google's API terms prohibit using the data to train their models. To keep text on-device too, route those features through your own local endpoint. See the technology page for the full breakdown.

The AI uses whatever context you give it. Spend two minutes filling out Your Context (your role, what you care about) and add a reference document — product positioning, pricing notes, objection handlers. It turns generic summaries into briefings that actually understand your work. Setup guide with examples →