DeepYardDeepYard
C

CAST

Case-driven framework that learns from execution trajectories to optimize LLM tool use

Open SourceFree

About

CAST is a research framework that treats agent execution trajectories as structured cases to improve LLM tool use. It extracts signals from past execution cases to optimize reasoning strategies and enhance structural validity in agent workflows. Designed for researchers exploring systematic approaches to agent learning from experience, it offers a novel case-based reasoning approach to improving tool selection and execution patterns.

Details

Language
Patterns

Tags

frameworktool-useopen-sourceautonomousevaluation