jevbooks

← All projects

typesafe-computer-use

awlevin/typesafe-computer-use

macOS computer use via OCR and bounded Jev action selection.

Agent Decisions99%Runner-up: Classification & Routing

What it is

A macOS computer-use loop that drives a Mac toward a plain-English goal. It reads the screen deterministically and asks a small classifier for the next action, avoiding screenshot calls to large models.

How it uses Jev

Each step sends one TypeSafe request with three Choices (four with off-screen controls): kind (click_item, use_browser, type_text, scroll…), item for click_item, site for use_browser, offscreen for press_offscreen. The returned choice becomes a deterministic action, then the loop waits and captures again.

Primitives:choice

Technique worth stealing

Split the decision into mutually exclusive questions so screen noise stays out of the action choice; overlapping options read as doubt.

Try it

uv run clicker "open the Playground" (dry run) or add --act to drive the machine.

View on GitHub

judged by Jevjev-1.13.0

Evidence

Each line is one question put to Jev about the README. ≥ 0.60 reads as yes, ≤ 0.40 as no; in between Jev is not making a call.

  • Jev-centricyes0.67
  • Shows a System One patternyes0.86
  • Handles uncertaintyno0.07
  • Measuredunclear0.56
  • Runnableyes0.73
  • Worth recommendingno0.38
  • Model replicaunclear0.56
  • Problem scopescore on a 0–2 scale0.79
  • About Jevyes0.98

Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 20 Sept 2026.