Sayso

Dictating into any Windows program needed a push-to-talk tool that types what was said, learns the speaker's vocabulary, and stays out of the way.

Period

2026-06 to present

Status

Built, in use

Stack

Rust, Tauri v2, React 19, TypeScript

Ownership

Sole builder

Problem

Hold a key, speak, release, and the words appear in the program that has focus. That is the whole product, and the hard parts are underneath it: typing into any program without corrupting the text, staying small enough to leave running all day, and getting names and project words right without the speaker correcting them each time.

Role

Kavish wrote the specs, chose the rewrite from Python to Rust, reviewed every change and uses it daily. Implementation was drafted by AI coding agents from his specs, under his review and tests.

Built

A rewrite in Rust

The first version was a Python app, started June 2026. It was retired in October 2026 for a Rust rewrite on the same stack as Cokpit: a Tauri v2 shell with a small React flyout, cloud speech-to-text, and a per-user installer tested on a clean virtual machine.

Typing that does not corrupt

Text is typed one character per input call. Batching characters corrupted the text in some programs, so the pacing is deliberate and tested.

A context engine

A vocabulary pool builds itself from what he dictates and keeps, with hints chosen by the program in front. A daily background review checks candidate words against kept recordings. Real words are never replaced automatically.

Result

Installed on his PC in October 2026 and in daily use; 132 Rust tests; about 4 MB committed and 0% CPU while idle. A 28-fix batch from an independent review was merged before install.

Collaborators

AI coding agents (Claude Code, Codex): drafted implementation from specs and plans he wrote, under his review and tests. No other contributors.

What stays private

Dictated text and the speech API key are never logged; recordings stay on his machine.