In-Meeting Notetaker.
On-Device Audio. One Shortcut.
- ✦ Every speaker identified — 99% accurate, recognized next meeting from a voiceprint that stays on your Mac benchmark
- ✦ No bot joins the call — captured at the OS level, so nothing appears in the participant list
- ✦ Action items and summaries — refined as you talk, through the default proxy or an endpoint you choose
- ✦ Keyboard-first — reachable from any app, no hunting through windows mid-meeting
Free meeting recordings and summaries. No credit card or email needed.
Apple Silicon · GDPR & CCPA · Fit it to your company's data policy →
References
Action Items
Summary
Transcript & Notes Generated Live
- Action items are associated with a labeled voice during the call.
- A full-accuracy transcription model runs in streaming mode; summary items with key points are generated on speaker changes.
- The final high-quality transcript is available seconds after call completion.
- Add reference documents before a meeting; the relevant passages come up when the conversation reaches them.
No Blending Between Individual Speakers
- MimicScribe uses signals from Nemotron 3’s diarizer and Pyannote Community 1, along with grammar clues, to perform speaker separation and determine the best sections of speech for speaker profile saving.
- AEC (acoustic echo cancellation) is run on both the mic and system channels to remove artifacts created by playing system audio over speakers or hearing your own voice in a participant’s audio.
Brightwave Reporting Discovery
Sam, you’re the one building these — walk me through what month-end actually looks like for you, step by step.
Sure. On the first I start exporting — GA4, Google Ads, Meta, LinkedIn — each one its own dashboard, one platform at a time.
Got it.
Then everything goes into one master sheet and I start reconciling. The platforms never agree on conversions, so I’m hand-matching attribution windows row by row.
And every client wants their numbers framed a little differently — it’s never quite the same report twice.
Right, so I rebuild the same charts four or five times. By the time the deck’s ready we’re already a week into the month.
And the renewals coming up next quarter — how many of those hinge on getting these out faster?
Labels on Voices
- The speech segments best representing the voice are found for profile enrollment and the profile is automatically given more samples as more meetings are added.
- A profile may be represented by multiple centroids so the same speaker in different room environments can be modeled without contamination of a single average.
- Only 20 seconds of clean speech is required to create a profile.
Fewer Words Lost
- It uses a custom harness that places encode windows in ways that optimize word recovery in tricky spots: overlapping speech, language code-switching, etc.
Undo History Ledgers
- All changes are recorded so a meeting or person can be restored to its original capture; MCP edits are undoable too.
- Log entries if transcript data is accessed or sent over the network; controls to allow and disallow access, including a "master switch".
Crash Recovery & Performance Tuning
- WAV files are saved to disk during recording; an entry in the sidebar allows you to transcribe them in the event of a failure.
- Memory is kept as flat as possible with minimal end-meeting peaking so even long meetings complete quickly and don’t cause stability issues. CPU use is about 30% of a single core.
Accurate LLM Processing
- The summary items are fed to the end-meeting prompt and there’s a bookmark system so you can ensure the model catches briefly mentioned topics.
- Tuning has been done to improve action item and summary completeness regardless of whether you’re using MCP, the OpenAI-compatible endpoint, or the bundled proxy.
- Different prompt sets are used based on model capabilities: on-device models, like Gemma 4 E12B, use lightweight prompts to remain responsive, while cloud models use richer prompts to handle date and time parsing.
Brightwave Reporting Discovery
Each Brightwave client report takes most of a week to hand-build across five ad platforms, and four renewals next quarter hinge on fixing it. The proposed build: scheduled API pulls into one store, live per-client dashboards, and an ask-anything agent — all on their own stack.
Action Items
- Mateo (You) — Send a scoped plan + sample dashboard for one client, built on their own data
Thu, Oct 22nd (by Friday) - Sam Rivera — Pull every platform they report on + the metrics each client cares about
Add to RemindersPain Points
Desired Outcome
Decision Makers & Approval Path
Open Questions
Wait, say that again. I’ll start recording.
Use the keyboard shortcut to trigger transcription only when needed.
Or let it start on its own.
Turn on Auto-Record and calls start recording when they begin.
Capture Presentations
- When turned on, snapshots of the shared window are captured as the meeting progresses; repeats are skipped.
- Faces are blurred before anything is saved.
- Pictures are placed inline in the final transcript, exported to Obsidian, and passed to MCP.
Connected to Your Tools
Ask Claude, Gemini, or Codex about any meeting, or find it in Spotlight. 94% of test searches put the right meeting in the top three.
> Add today's action items to my Linear inbox.
⏺ mimicscribe · get_action_items
Added 3 issues from "Weekly standup":
· Ship the export fix (Sam, Fri)
· Review the onboarding copy (you, Thu)
· Book the customer interview (Priya, next week)
Connect Your AI Agent
Claude
Gemini
CodexClaude, Gemini or Codex can search, read and update your meetings.
Set up integrations →Notes In Your Vault
Obsidian
LogseqAutomatic Markdown notes in your Obsidian vault or Logseq graph. No agent required, and nothing has to leave your Mac.
Set up the notes folder →Native to Your Mac
Meetings show up in Spotlight and plug into Shortcuts and Siri. It's built on Apple's App Intents, so there's no scripting.
Set up shortcuts →Free to Start
Recording, transcription and speaker separation are free and run on your Mac. Pro is a one-time purchase. Cloud AI is optional.
Recording and transcription- Speaker separation and voice profiles
- Voice typing
- Search across meetings
- AI meeting summaries for light use
- 120 in-meeting briefings a week
- MCP server for AI agents 100 calls / week
- Auto-export Markdown 5 meetings / week
- OpenAI-compatible endpoint 15 summaries / week
Everything in Free, no limits on
- MCP server for AI agentsClaude, Cursor or any MCP client
- Auto-export MarkdownObsidian, Logseq or any folder
- AI on any OpenAI-compatible endpointOllama, LM Studio, Microsoft Foundry, Bedrock
- Yours forever, all 1.x updates
- Activate once, works offline for good
Unlimited summaries, briefings, follow-ups and file imports on our Gemini proxy. Includes Pro while subscribed.
Requires macOS 15+ and Apple Silicon · v1.0.0-rc.40 · secure checkout by Stripe · 30-day money-back guarantee
No account, no usage analytics. Cloud AI has fair-use limits that heavy use doesn't reach.
Questions about plans or billing? support@mimicscribe.app
On-device by Default. You Pick the Provider.
All audio is processed on your Mac. AI is opt-in: a local model, your company's provider, or our open-source Gemini proxy.
Using it at work? Read the company-policy guide →
Bring Your Own Endpoint
Ollama and LM Studio locally, your company's Foundry or Bedrock, or our Gemini proxy. Swap it anytime in Settings.
On-device Per Meeting
Turn it on for every meeting or just one. Transcription and speakers keep running on your Mac, and no text goes to a remote AI.
No Account
No sign-in, no email, no usage analytics. Pro is a license key, not an account.
See Every Request
Settings → Network Log lists each request live: what it was for
and where it went. Check with nettop too.
Common Questions
Compatibility, privacy, permissions and requirements.
Does it work with Zoom, Teams, and other apps?
Yes. MimicScribe captures system audio at the OS level — no meeting bot, plugin, or attendee-list badge. It works with Zoom, Teams, Google Meet, Slack huddles, or any app that plays sound, and can start and stop recording automatically when a call begins.
Do I need to tell people I'm recording?
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 participants; that's your call. Best practice: tell attendees at the top of the call, and stop if anyone objects. In a regulated industry, check with legal or compliance first.
What permissions does MimicScribe need?
- Microphone — required for all voice features.
- System Audio Recording — captures audio from video calls. Audio only — no screen capture or video.
- Accessibility optional — voice editing, plus recognizing meeting sites in a browser for automatic recording. Skip it and meeting recording still works, including automatic recording for Teams, Slack, and Zoom.
Does it handle other languages and technical terms?
Yes — 25 languages, detected automatically. The transcript and the summary both come out in the language spoken. Technical terms and proper nouns work without a custom dictionary. See the languages page for the list.
Will MimicScribe slow down my Mac?
No. It sits at about 145 MB in the menu bar, and a meeting costs roughly a third of one CPU core, because the heavy ML work runs on the Neural Engine. Runs on any Apple Silicon Mac with macOS 15 or later, including a base M1 with 8 GB. Setup downloads about 1.2 GB of models. There's no Intel build. See full performance details →
How does voice typing work?
Hold a shortcut to dictate cleaned-up text at your cursor in any app, or select text and speak an instruction to rewrite, translate, or change tone in place. Like meetings, speech recognition runs on your Mac. Dictation → · Voice editing →
Anything else, email support@mimicscribe.app and I'll get back to you.