Cua Driver jev-use
A public-preview Cua Driver recipe for bounded Jev browser actions with independent verification.
What Jev decides
Which supplied action candidate to take next, or whether to reobserve or abstain.
Equivalent Python and TypeScript loops observe through Cua Driver, construct immutable snapshot-bound candidates, let Jev select one ID, validate it, execute through Driver, and verify the fixture state separately.
- Jev role
- Choose one immutable browser action candidate while Cua Driver owns observation, execution, permissions, and verification.
- Input
- The task goal, compact current observation, capture identity when applicable, bounded history, and at most 32 candidate IDs with descriptions.
- Output
- One allowlisted candidate ID, confidence, and the probability distribution over candidates.
- Decision frequency
- Once per step in the bounded browser decision loop.
- Fallback
- Reobserve or abstain, reject stale and malformed candidates, and never retry an uncertain action blindly; verify success through independent fixture state.
How it works
Overview
Cua Driver jev-use is a public-preview example inside the Cua repository rather than a standalone product. It keeps TypeSafe-specific provider code above the driver and demonstrates a narrow decide-act-verify boundary in both Python and TypeScript.
Candidates are built from fresh browser semantics or validated visual regions and always include reobserve and abstain. Jev sees compact observations and allowlisted IDs, not screenshot bytes or executable tool arguments. The credential-free mock proves the Driver lifecycle, action loop, stale-reference handling, and postcondition checker; live Jev verification is optional and must be recorded separately.
- Language
- Python / TypeScript
- License
- MIT
- Author
- Cua contributors
Sources
Related projects
jev-ultrafast
A browser agent where Jev selects an operation and a compatible page element from a live indexed action space.
What Jev decides
Which supported browser operation to run, and which compatible visible element to target
Jev Review
A code-review pipeline assembled from small typed judgments instead of one long reviewer prompt.
What Jev decides
What is risky, which files and evidence matter, what could fail, how severe it is, and who should review it.
Jev Voice Browser
A browser that starts resolving voice commands while the user is still speaking.
What Jev decides
What the user intends, which page target it applies to, whether the command is complete, and whether it is destructive.