name: gepa
display_name: GEPA
organization: "GEPA"
type: test_time_search
family: evolutionary_method
summary: "Reflective mutation over execution feedback, with Pareto-frontier candidate selection."
source: "GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learning (Agrawal et al., ICLR 2026) — github.com/gepa-ai/gepa"
tags: [reflective-mutation, pareto, feedback-driven, prompt-evolution, sample-efficient]
license: MIT
controller: galapagos.scaffolds.gepa.scaffold.GepaScaffold
components:
  population: {kind: pareto_candidate_pool}
  selection_policy: {kind: pareto_frontier_sampling}
  prompt_builder: {kind: gepa_reflection}
  proposer: {kind: reflective_rewrite}
  evaluator: {kind: task}
  memory: {kind: none}