Proposers

Proposers are strategies that generate new candidate programs during optimization. GEPA provides two main proposer types:

Reflective Mutation Proposer

The primary proposer that uses LLM-based reflection to improve candidates based on execution feedback.

• ReflectiveMutationProposer - Main reflective mutation proposer

Merge Proposer

A proposer that combines successful candidates from the Pareto frontier.

• MergeProposer - Merge-based candidate proposer

Base Classes and Protocols

• CandidateProposal - Data class for candidate proposals
• ProposeNewCandidate - Protocol for proposer strategies
• Signature - Base class for LLM prompt signatures
• LanguageModel - Protocol for language models