No Bot in the Meeting: How MinuteAI's Chrome Extension Transcribes Google Meet and Teams Calls
MinuteAI's Chrome Extension captures your own audio and streams it to localhost — no bot joins the call. Here's the architecture, and why 2026's platform crackdown on meeting bots makes that design look prescient.
The Bot That Joins Before You Do
Open a Google Meet or Teams call with Otter, Fireflies, or Grain connected to your calendar, and something joins the meeting alongside you: a bot, with its own name and its own tile, announced to everyone in the room. “Otter Bot has joined the meeting.” Whoever’s on the other end sees it, whether they expected a recording or not.
MinuteAI’s Chrome Extension does the same job — turning a Google Meet or Teams call into a searchable transcript — without that step. It doesn’t join the meeting. It captures the audio playing in your own browser tab, streams it to a small server running on your own Mac, and processes it there. No second participant shows up in the call. No third-party server ever receives the audio at all.
That distinction has always mattered for privacy. In 2026, it’s started to matter for whether the bot is allowed in the room at all.
How It Actually Works
The mechanics are unglamorous, which is part of the point. The Chrome Extension captures audio from the active browser tab — the same audio you’re already hearing as a participant — and streams it to localhost:8085, a server that MinuteAI’s desktop app runs on your own machine while it’s open. The app picks up the stream, runs it through an on-device transcription engine, and writes the transcript straight into your MinuteAI library. The extension never talks to any server that isn’t the one running on your own Mac.
That local server isn’t reachable from outside your machine — it only accepts connections from your own browser, not from MinuteAI’s infrastructure or anyone else’s. There’s no upload step for the audio to leak from, because there’s no upload step at all.
Three transcription engines sit behind that pipeline, and which one runs is a genuine trade-off rather than a single default dressed up as a choice: WhisperKit covers 99 languages and is the most accurate of the three but the slowest — roughly 8-12 minutes to process a 30-minute meeting; FluidAudio covers 55 languages at up to 50x the speed, with a small accuracy cost; Apple’s built-in Speech Analyzer covers 45+ languages and runs fast using macOS’s native speech recognition, at the expense of accuracy on technical vocabulary. Nothing about that trade-off requires a network call in either direction — all three run locally.
For Microsoft Teams specifically, the extension only works with the browser (teams.microsoft.com), not the desktop client, because it can only capture audio from a browser tab. That’s a real limitation worth naming, not a footnote to bury — a design that captures tab audio can’t reach into a native app’s audio pipeline, so Teams users have to be in the web client for the extension to see the call at all.
What “No Bot” Actually Buys You
The practical difference between joining as a bot and capturing your own tab audio isn’t cosmetic. A bot needs to be let into the meeting — in a hosted call, someone with admin rights either approves it explicitly or the meeting’s settings allow third-party participants by default. Capturing your own tab’s audio requires none of that, because nothing new is entering the call. You’re already an authorized participant; you’re just also running a local recorder.
That also means no visible notification to the rest of the call that a third-party recording tool has been added — which cuts both ways. It removes the “notification fatigue” that visible bots create, but it also means other participants don’t get a passive signal that the conversation is being recorded, so any disclosure obligation still falls on the person running the extension, not on the tool.
Why 2026 Is a Bad Year to Be a Bot
The market context here isn’t hypothetical — it’s actively unfolding this year. Google Meet’s March 2026 update started flagging third-party notetaker bots as a “potential risk” by default, requiring the meeting host to manually approve each one before it can join. Microsoft published a comparable move in notice MC1251206 in March 2026: detected bots now get routed into a “Suspected threats” lobby labeled “Unverified,” with general availability rolling out through June 2026. Both changes point the same direction — platform vendors are making the bot-joins-a-call pattern something a host has to actively let through, not something that happens by default.
The legal pressure is compounding the platform pressure. GDPR requires affirmative consent before recording begins, and a visible bot in the participant list isn’t treated as sufficient disclosure in any jurisdiction — presence isn’t consent. That gap has turned into active litigation: Fireflies.ai is facing suits under Illinois’ BIPA, and Otter.ai is facing suits under the federal ECPA and California’s CIPA, both testing whether the person who enabled the bot is on the hook for getting consent from everyone else on the call. Enterprise legal and IT teams have started responding by banning cloud recording bots outright at the admin level — the bot doesn’t get denied per-meeting, it never gets the option to ask.
None of that pressure touches a tool that never joins the call as a separate entity in the first place. A browser extension capturing your own tab’s audio and processing it on your own machine isn’t a bot a host has to approve, and it isn’t a cloud vendor a legal team has to vet for data residency or third-country transfer. It’s closer, in practice, to recording a call with your own microphone — which is exactly the analogy MinuteAI’s own product documentation draws.
The Trade-Off That’s Still Real
None of this makes the extension a free pass on consent law. One-party consent states let you record a conversation you’re part of without telling anyone else; two-party consent states — California and Florida among them — require every participant’s agreement regardless of whether a bot shows up in the roster or the recording happens silently on your own machine. Running MinuteAI’s extension instead of a cloud bot changes who else has access to the recording; it doesn’t change whether you were legally required to ask first. That responsibility sits with whoever clicks “record,” same as it always did.
Where This Fits
We’ve written before about MinuteAI’s on-device transcription pipeline — the model manager, the memory constraints, the reasons none of it touches a server. The Chrome Extension is the same discipline applied to a different capture point: instead of a voice memo or an imported recording, it’s a live meeting tab, and instead of an upload that never happens, it’s a bot that never joins. Same non-negotiable, applied to a moment — the middle of a call — where most competing tools still assume a round trip through someone else’s infrastructure is unavoidable.
MinuteAI’s Chrome Extension is free, with no usage limits on the browser capture itself. Learn more at getminute.app, or reach out at hello@aitytech.com if you want to talk through how it fits a specific meeting workflow.
See Our Work
From MinuteAI to AgentKits — explore the products and projects we've shipped.
View PortfolioRelated Articles
Apple and Google Just Started Transcribing Calls for Free. Neither One Touches a Zoom Tab.
iOS 26 and Google's Pixel Recorder now do on-device call transcription and summarization for free. Here's the specific boundary neither platform crosses — and why that's exactly where MinuteAI's Chrome Extension operates.
GuidesJapan's ¥12 Trillion Legacy-System Warning Isn't a COBOL-Skills Story. It's an Encoding Story.
METI's 2025 digital cliff warning gets read as workforce attrition and rip-and-replace math. The failure mode that actually corrupts migrations first is smaller and easier to miss: EBCDIC and Shift-JIS don't even agree on whether letters or numbers sort first. Why Legacy Dragon treats character encoding as a parsing-layer concern, not a pre-processing step bolted on later.
GuidesNo Pricing Page, No API Meter: The Economics Behind PrivateAI's Free Tools
Cloud AI is priced by the token because every query costs the vendor real compute. On-device tools don't have that bill. Here's what that structural difference actually buys — and doesn't — for a product like PrivateAI.