savka777/jev-use

★ 71⑂ 7

Say it, and your Mac does it. A computer-use harness on Jev that reads the screen through Accessibility. Fast, no vision model

About savka777/jev-use

savka777/jev-use is an open-source project on GitHub, mainly written in Swift. Say it, and your Mac does it. A computer-use harness on Jev that reads the screen through Accessibility. Fast, no vision model It currently holds 71 stars and 7 forks with 1 open issues, and was last pushed on 2026-09-19 (repository created 2026-09-19).

Project Overview

AI Homed tracks it on the Today's Trending board, currently at rank #97 with 0 new stars today.

GitHub Repository Details

Repository savka777/jev-use · default branch main · size 195 KB · watchers 0 · source: GitHub REST API and repository README

README

https://github.com/savka777/jev-use/blob/HEAD/jev-use

jev-use

Voice and typed computer use for macOS. You say what you want. Jev picks the next on-screen action. macOS performs it. No screenshots: the app reads the screen through the Accessibility tree.

Quick start

Requires macOS 14.2+ and Xcode. No dependencies.

bash build.sh
open "$HOME/Applications/Desktop Voice.app"

In setup: save your TypeSafe API key (stored in the Keychain), then allow Accessibility, microphone and speech.

Hold Control–Option–Space, speak, release. Escape cancels. Or type a command in the widget, or from a shell: scripts/say.sh "Open Finder".

Examples

How it works

One loop, about 0.3–1.5 s per step:

1. Read. Walk the front app's Accessibility tree (~120 ms). Every element describes itself: what it is, its name, its value, where it sits, what it can do. No per-app code. 2. Choose. One request to Jev (jev-latest): the goal, the numbered targets, the last ten actions and their effects. Jev selects an operation and a target. It never generates free text; typed text is a span of your sentence. 3. Act. Press, select, type, menu, key, scroll, open, arrange windows. 4. Check. Read the screen again. Report the real effect. Repeat until DONE, BLOCKED or WAIT.

Low-confidence and destructive picks stop and ask instead of acting.

What is sent

To https://api.typesafe.ai/v1/systemone: your command, the app and window names, the on-screen targets with their labels and values, and recent actions. Secure text fields are excluded. No screenshots. Speech uses Apple Speech.

Everything is logged locally: log show --predicate 'subsystem == "local.jev-use"' --last 10m --info

Develop

swift test        # 10 tests
bash build.sh     # quit the app first

GitHub Stars & Activity

71Stars
7Forks
1Open issues
SwiftLanguage

GitHub Popularity

GitHub stars71
Forks7
Open issues1
Primary languageSwift
LicenseMIT
Stars gained today0
Created2026-09-19
Last pushed2026-09-19

Trending History

Daily boardrank #97 · ▲ 0 stars

Related AI Projects

1

Beingpax / VoiceInk

Swift★ 6,492⑂ 924▲ 24 stars
2

mattt / iMCP

Swift★ 1,634⑂ 120▲ 11 stars
3

haplollc / ThinkingOrbs

Swift★ 77⑂ 5
4

affaan-m / ECC

JavaScript★ 263,994⑂ 39,485▲ 826 stars
5

tensorflow / tensorflow

C++★ 200,221⑂ 77,033▲ 30 stars
6

Significant-Gravitas / AutoGPT

Python★ 187,469⑂ 46,006▲ 27 stars
7

anthropics / claude-code

TypeScript★ 147,260⑂ 24,085▲ 419 stars
8

ggml-org / llama.cpp

C++★ 129,000⑂ 23,514▲ 136 stars

More AI Rankings