Run Low-Cost, Deterministic Computer Use with TypeSafe
A beta Python toolkit that drives macOS (and experimentally Windows) from plain-English goals by combining deterministic screen reading with a small classifier.
What you get from it
What it is
typesafe-computer-use is an open-source computer-use loop for Python 3.12+. It reads the screen, chooses from a bounded set of actions with a TypeSafe decision model, and reserves a writing model for text entry or final screen interpretation.
Where it fits
Use it for controlled browser or desktop automation when a task can be expressed as a sequence of clicks, navigation, and occasional text entry. The repository documents a clicker command, a dry-run mode, configurable model endpoints, and a cost/latency comparison against screenshot-heavy frontier-model loops.
Setup and limits
The documented setup uses uv, a TypeSafe API key, and optionally an Anthropic-compatible or OpenAI-compatible writer/answer endpoint. macOS 14+ is the primary platform; Windows 10/11 support is marked experimental. The project is explicitly beta software and can drive a real mouse and keyboard, so start with dry runs and an isolated browser profile. Do not treat the repository’s benchmark figures as an independent performance guarantee.
The code is MIT-licensed. Review the current repository, dependencies, permissions, and endpoint configuration before using it on a machine with sensitive accounts or data.
Discussion
Share practical experience, questions, or warnings with the community.
Sign in to join the discussion and vote on comments.
Sign in