Kennan
About

A meeting AI that doesn’t phone home.

I’m Luiz, the person behind Kennan. Here’s the short version of why I built it.

I kept losing the thread in conversations I cared about. A salary review where I knew the right number but couldn’t anchor. A hard performance conversation where I forgot the antidote to defensiveness until I was already in it. The moves were knowable. They’re named, they’re researched, they’re in the books. But the moment was gone before I could reach for them.

The tools that exist for meetings are built for after the call. Otter, Granola, Krisp: they hand you a transcript or a summary when it’s already too late to use it. The handful of real-time tools route your audio through someone else’s server. I wanted something that helps in the moment without the conversation ever leaving my Mac.

Kennan is that. Transcription on-device. Optional fully-local language model. Bring-your-own-key for OpenAI, Anthropic, or Gemini if you want a frontier model: your Mac talks to the provider directly, with no server in between. No account, no email signup, no subscription. The HUD is invisible to screen-share and screen-recording. The other side never knows it’s there.

It’s built in Swift, runs on macOS 15+, and the research library on the site is the spec. Every suggestion the assistant emits traces back to a named framework with a citation. If you want to see how, that’s the page to read.