rsi-gear
A research framework for improving LLM agent configurations through task experience and reproducible evaluation
SITE CAPTUREBased on information published on the service's website. Selection and review policy
WHAT IT DOES
What does it do?
GEAR lets agents refine their reasoning, planning, and actions within individual tasks and identify recurring failures and effective strategies across tasks. It evaluates proposed improvements for use in later tasks and allows components of the evolutionary search algorithm to be replaced. It also provides materials for examining experiment records and evaluation evidence.
KEY FEATURES
Key features
- Improves agent reasoning, planning, and actions using task feedback
- Finds recurring failures and effective strategies across tasks
- Evaluates improvements before applying them to subsequent tasks
- Allows experimentation by replacing components of the evolutionary search algorithm