You talk, Sumi types — but not what you actually said. It cleans up filler words and grammar, then adjusts tone for whatever app you're in. Gmail gets a professional version, Slack gets something short. Meeting mode records calls in the background. Edit by voice: select text, say what to change. 100+ languages. Runs on your machine. Free, GPLv3. macOS + Windows.
No reviews yetBe the first to leave a review for Sumi
Viktor.comAn AI coworker that actually does the work
Promoted
Maker
📌
I'm a developer in Taiwan. I built Sumi because I run multiple Claude Code agents in parallel and typing instructions all day was the actual bottleneck. macOS dictation just types what you say, filler words and all. The paid alternatives charge $10-15/month and send audio to the cloud.
The feature I use most is the context-aware polish. I say something sloppy and it comes out clean, with the tone matching whatever app I'm in. No cloud round-trip.
I also quantized the Qwen3-ASR model myself and wrote the inference pipeline in Rust. It handles accented speech and dialects better than Whisper in my testing.
Happy to answer any questions about the architecture or the tool itself.