#6060th of 82 in Agent frameworks and SDKs
gepa
Python library that evolves prompts and other text artifacts using LLM reflection
- Stars
- 7.2k
- License
- MIT
- Last commit
- Oct 2026
- Last release
- Jul 2026
- Language
- Jupyter Notebook
3,136 on 4 Apr 20267,229 on 31 Oct 2026
From GitHub's daily star counts.
Checked on GitHub.
Overview
GEPA (Genetic-Pareto) optimizes text parameters such as prompts, code and agent configurations against a metric you supply. An LLM reads full execution traces to diagnose failures, proposes mutations, and keeps a Pareto frontier of candidates. It exposes gepa.optimize, a general optimize_anything API, adapters for DSPy, MCP, RAG and LangChain, and is also available as dspy.GEPA.
Who it is for: Engineers tuning prompts, agents and configs against a measurable metric
Strengths
- Reads full execution traces, not just a scalar score, to guide each mutation
- README reports 100-500 evaluations versus 5,000-25,000+ for GRPO
- Built-in adapters for DSPy, MCP, generic RAG, LangChain and TerminalBench
- Integrated in DSPy, MLflow, Opik, Pydantic AI and Google ADK
Weaknesses
- Needs a measurable evaluator and a dataset or metric to optimize against
- Reflection step depends on a strong LLM (examples use openai/gpt-5), adding API cost
- Optimizes text artifacts only; no weight fine-tuning
- Benchmark claims in the README come from the authors and partners
What it needs
- no GPU
- Needs LLM API for task and reflection models (e.g. OpenAI)
- Models: OpenAI-style provider/model strings, any LLM via adapter, DSPy programs, LangChain pipelines
Quick answers
Is gepa open source?
Yes. Its code is on GitHub under the MIT license.
Does gepa need a GPU?
No, it runs without one.
Also in Agent frameworks and SDKs
See all 82| Rank | Project | Score |
|---|---|---|
| 1 | ppt-masterAgent skill that turns documents into natively editable PowerPoint decks | 95 out of 100 |
| 2 | openai-agents-pythonPython SDK for multi-agent workflows with handoffs, guardrails and tracing | 90 out of 100 |
| 3 | langchainPython framework for building LLM agents and apps from interoperable components | 89 out of 100 |
| #4 | browser-usePython library that lets an LLM agent control a web browser | 89 out of 100 |
| #5 | tradingagentsMulti-agent LLM framework that simulates a trading firm's analysis process | 89 out of 100 |