CalibratedDecisions.

Repo · Agents & browsers

JevHarness

LLM-written Jev harnesses, with reward reflection over the full trajectory.

Open github.com ↗

From the repository

LLM-authored task-specific Jev harnesses with optional full-trajectory reward reflection and GEPA evolution.

How builders describe it

LLM-authored task-specific Jev harnesses with optional full-trajectory reward reflection and GEPA evolution.
A Python framework where an authoring LLM writes a task-specific Jev harness, then optionally evolves it using reward reflection and GEPA over full trajectories.

The decision Jev makes

Pick the next click, field or navigation from the legal actions on screen.

Where it fits

Browser agents need to choose an action on every step. With Jev each step is a choice from the actions actually available on the page, which keeps agents fast and on the rails. All 299 agents & browsers projects →

Related projects