wy-coliney/jev-browser-use

What it solves

Jev Browser Use accelerates browser automation by splitting the workload between a fast, specialized model (Jev) and a more capable general-purpose agent (Codex). It eliminates the need for a high-cost, high-latency model to handle every single click, scroll, or toggle, which typically slows down browser-based AI agents.

How it works

The project implements a handoff mechanism where Jev handles the repetitive navigation and control actions based on accessibility text. Codex remains in charge of high-level planning, text input, visual interpretation of screenshots, and final verification of the results. This creates a loop where Jev executes actions and Codex verifies the outcome, avoiding a new host-model turn for every single interaction.

Who it’s for

Users of Codex and other AI agents (like Claude Code) who perform repetitive browser tasks—such as navigating dashboards, managing settings, or reviewing reports—and want to reduce latency and API costs.

Highlights

• Speedup: Claims a 5–10× increase in browser operation speed in specific workflows.

• Cost Efficiency: Uses Jev's low-cost API to handle repetitive decisions, reserving expensive models for complex reasoning.

• Integration: Works as a Codex Skill or plugin, utilizing existing browser connections without requiring additional drivers or npm dependencies.

• Hybrid Approach: Combines accessibility-text-based navigation (Jev) with visual judgment (Codex).