SylphAI-Inc/AdalFlow

Implements gradient-based prompt optimization through LLM-AutoDiff and few-shot learning frameworks, enabling zero-shot and few-shot auto-optimization without manual tuning. Built with PyTorch-like compositional components (agents, RAG pipelines, retrievers) that abstract model differences, supporting seamless switching between LLM providers via configuration. Includes native tracing, human-in-the-loop controls, and MLflow integration for production monitoring without external dependencies.

4,068 stars. Available on PyPI.