Overview & Result
Traditional browser agents pause 3 to 10 seconds per step waiting for multimodal LLMs to output next clicks. Jev-RA decouples visual/DOM parsing from action selection, allowing Jev to pick the next DOM interaction in ~300ms, making browser agents feel interactive.
How Jev fits in the loop
- Capture page DOM snapshot and extract interactable candidate elements.
- Formulate the next step decision as a choice among actionable elements.
- Jev evaluates the candidate set against the immediate sub-goal.
- Playwright executes the click/type action and observes the resulting state.
How to reproduce
- Clone brnyxx/jev-ra and configure Playwright headless browser.
- Define a targeted web exploration task (e.g. item lookup or link verification).
- Execute the reactive agent loop and track step execution timing and decision logs.
Why this build matters
Brings reactive latency down to sub-second speeds in browser agents, making practical automation viable.
Reported performance
Reported by authorLatency: ~300 ms per step
Author-reported step latency across standard web task benchmarks.
Limitations
- Requires reliable DOM selector generation from the underlying page state.
- Complex multi-page state transitions require an external planner to maintain macro goals.
- Network delays on heavy sites can outweigh decision latency.