DeepYardDeepYard
E

EvoTool

Evolutionary framework that self-optimizes tool-use policies for long-horizon LLM agent tasks

Open SourceFree

About

Research framework using evolutionary algorithms to optimize how LLM agents select and use tools across complex, multi-step tasks. Employs blame-aware mutation to identify which tool choices caused failures and diversity-aware selection to maintain exploration. Addresses the credit assignment problem in long-horizon agent trajectories where traditional fine-tuning struggles with delayed feedback signals.

Details

Language
Patterns

Tags

frameworktool-useautonomousopen-sourceevaluationpython