MCTS Agent
Use Jev judgments inside a Monte Carlo tree search agent.
What Jev decides
Action prior probabilities, state value scores, branching width, and whether the goal is complete.
Gemini proposes actions; Jev supplies action priors, state scores, branching choices, and completion judgments for the search loop.
- Jev role
- Evaluate candidate actions and states during search.
- Input
- The goal, current state, and Gemini-proposed actions or hypothetical states.
- Output
- Typed Choice, Score, and Noul judgments consumed by the search algorithm.
- Decision frequency
- Across planning and reassessment steps of the MCTS loop.
- Fallback
- The agent executes only the selected immediate action and grounds later decisions in observed results.
How it works
Overview
This experimental agent combines Gemini action generation with a discriminative MCTS planner. Jev Choice assigns priors and branching widths, Score evaluates hypothetical states, and Noul checks completion. The current README says low early scores do not remove proposed paths; execution observes the real workspace before the next planning step.
- Language
- Python
- Author
- lhemerly
Sources
Related projects
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.
Distill
Route coding-agent work, reasoning effort, and context retention through bounded Jev decisions.
What Jev decides
Which configured model and effort should handle a call, what evidence deserves attention, whether another attempt is useful, and what context should be retained.
jev-model-router
Route Claude Code model tiers and reasoning effort with one parallel Jev classification.
What Jev decides
Which fast, balanced, or deep tier fits the work, how much reasoning it needs, and whether production risk requires escalation.