Lambert de beru

Kai for Chrome - Local meeting transcription with no account needed

by
Kai transcribes & summarizes your Meet, Zoom, and Teams calls 100% on-device. Press ⌘⇧K, get a clean summary + action items the moment the call ends. No account, no upload, no meeting bot in the room. First Chrome extension running Whisper Base fully locally via WebGPU + WASM. No cloud STT, no API call, no signup. Free forever. Built by Morgen as the entry point to Kai, our AI executive assistant.

Add a comment

Replies

Best
Lambert de beru
Good morning folks, We shipped this last week !! A free Chrome extension. Press ⌘⇧K on any Meet, Zoom, or Teams tab and Kai transcribes the call, then writes a clean summary + action items when you stop. 100% on-device. Whisper Base runs locally in your browser via WebGPU. No account. No upload. No bot in the call. Every meeting tool we tried wanted us to either pay per seat, sign up, send our audio to their servers, or drop a bot into the call. This one does none of that. It's free. It's going to stay free. It's also the first surface of Kai, the AI executive assistant we're building at Morgen. The Chrome extension is the entry point. Full Kai (email, calendar, triage) is on waitlist at hirekai.ai. If you try it and something feels off, tell me. Especially curious how it holds up on long calls and accents we haven't tested. → https://chromewebstore.google.co... Lambert (Kai team)
Jim Hartung

@lambertlcdb awesome job. already my fav extension!

Marco Ancona
@lambertlcdb great job Lambert and team. Simple and effective
Alaa Moucharrafie

@lambertlcdb "No account", "100% on-device", "It's going to stay free" -- *chef-kiss* 😚👌

Luigi Pederzani

Congrats on the launch!
PS: How to get a massage from Kai? Asking for a friend 😂

Vikram

Super impressive work getting Whisper Base running smoothly via WebGPU inside a Chrome extension! I've been experimenting with WASM implementations for local ML, and memory management inside a long-running browser tab is usually a nightmare. Have you guys run into any browser heap limits or thermal throttling issues during heavy, multi-hour calls? Especially on older Intel machines or base-model M1 Macs where Chrome is already fighting for RAM with Zoom or Teams running in parallel. Would love to know how you optimized the audio buffer processing without crashing the page. good job team

Lambert de beru

@vikramp7470 Thanks! One thing that's actually the heart of your question: it's not pure WebGPU. When WebGPU isn't available (old Intel, base M1, machine already loaded), it automatically switches to a lighter backup mode. We really designed for those machines first.

The main idea on memory: it never grows with the length of the meeting. A 3-hour call uses about the same memory as a 3-minute one. The audio flows through a fixed-size "box" (around 1 minute of sound, reserved once at startup). Nothing piles up, so there's no leak that eventually crashes the tab. And the transcription runs in a separate lane, never on the page itself, so Meet/Zoom/Teams don't lag while it works.

On not crashing the page: if the machine can't keep up, the sound just waits in that box instead of stacking up forever. If it really overflows, we drop a small bit of the transcript instead of crashing. Same with heat: if the chip slows down, transcription slows down a little, that's it, nothing breaks.

Happy to go deeper on any of it!

André Jonas

Who needs Granola when you have stuff like this! Good job!

David Tedaldi

For the longest time, I've only had one password and ad blocker extension, but it took literally less than a day to get hooked on Kai's extension! Great work @lambertlcdb

Evak Chen

Congrats on the launch! This looks so cool, especially the "no account needed". But I'm also curious about if there's no sign-in, how does Kai handle meeting history? Is everything saved locally in the device?