Cua Pairs Jev with Its Driver for 5x Faster Computer Use
- Source
- Cua
- Date
1/ Fast Computer Use is now solved with @typesafeai Jev + Cua Driver. Available in development preview for macOS, Windows, and Linux. We call it jev-use. Draft #3943: https://t.co/J1frbEbZTQ
2/ Cua Driver + Jev playing 2048. In this measured 2048 run, Jev was 5x faster and about 1,000x cheaper than Astra. That result is specific to this run, not a general claim about either system. Video attachment: 30 seconds, silent, 1920 x 1080. The final frame shows Jev at 44.9 seconds and about $0.00108 API-equivalent cost, versus Astra at 294.9 seconds and about $1.45. Demo by @injaneity
3/ Jev from @typesafeai is the fast decision layer. Cua Driver observes and executes across macOS, Windows, and Linux. jev-use turns what Driver sees into bounded action candidates Jev can select, then Driver executes and verifies.
4/ For universal GUI parsing, we're introducing optional Cua Perception. It can use screenshot parsers such as OmniParser to produce typed visual regions when AX or DOM is incomplete. Cua Driver stays model and parser agnostic.
- In one measured 2048 run, Jev finished in 44.9 seconds at about $0.00108 API-equivalent cost versus Astra at 294.9 seconds and about $1.45. Cua says the result is specific to that run, not a general claim about either system.
- Optional Cua Perception can run screenshot parsers such as OmniParser to produce typed visual regions when accessibility or DOM data is incomplete, and is skipped when AX or DOM already provide enough structure. Cua Driver stays model and parser agnostic.
- The jev-use loop stays bounded: Driver observes, Perception parses if needed, the client builds candidates, Jev selects one ID, and Driver executes and verifies fresh state. It is in development preview on macOS, Windows and Linux.
- Draft PR #3943 lays foundations such as optional extensions, typed visual-region contracts, bounded captures and a fixture Rust worker protocol. Public parse and runtime wiring, an optional OmniParser install and Jev provider integration are still listed as next steps.
- LLM — A large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.
- AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
Shows a concrete architecture pattern for cutting computer-use latency and cost dramatically by reserving full calls for planning and using a cheap classifier for local, bounded decisions.
Checking sign-in…
Loading comments…