Topics / #self-evolving-agents

Self-evolving-agents projects on GitHub

AI agents that improve themselves through experience or optimization loops

As of , the most-starred Self-evolving-agents projects on GitHub are SkillOpt (18.5K stars), darwin-skill (6.4K stars) and hermes-agent-self-evolution (5.5K stars). GitHub.Locker tracks 13 Self-evolving-agents repositories with 40.2K stars combined. The fastest-growing on October 17 was Harness-RSI, with +46 stars in a day.

13Projects
40.2KStars in total
+122Stars in a day
Browse in the catalog →github.com/topics/self-evolving-agents ↗

Hot right now

Most stars gained in a day · Oct 17

+46

Harness-RSI

zgcagi/Harness-RSI · ★ 170

Companion materials for our survey of how execution experience becomes persistent harness state under a fixed base model. The survey covers update mechanisms, evolutionary search, continual PAC learning, evaluation, applications, and societal impact.

+27

SkillOpt

microsoft/SkillOpt · ★ 18.5K

A text-space optimizer that trains reusable natural-language skills for frozen LLM agents without touching model weights. The trained skill ships as a compact markdown file (typically 300–2,000 tokens) that adds zero inference-time model calls at deployment.

+14

Deepgraph

billion-token-one-task/Deepgraph · ★ 1K

DeepGraph is the autonomous research agent of JouleBeat, an open-source AI self-improvement (RSI) lab. It reads scientific literature, finds open questions, runs real experiments, and lets the results decide what is true.

+14

awesome-rsi

lobehub/awesome-rsi · ★ 766

Awesome RSI is a curated research map of Recursive Self-Improvement (RSI): a taxonomy of papers, frameworks, benchmarks, and safety work on AI systems that improve their own capabilities and the mechanisms that generate further improvements.

The all-time top

13 projects · by stars

  1. #1 · microsoft/SkillOpt

    SkillOpt

    A text-space optimizer that trains reusable natural-language skills for frozen LLM agents without touching model weights. The trained skill ships as a compact markdown file (typically 300–2,000 tokens) that adds zero inference-time model calls at deployment.

    18.5K
  2. #2 · alchaincyf/darwin-skill

    darwin-skill

    Darwin.skill is a system that evolves Claude Code Agent Skills through an automated evaluate-improve-test-keep-or-revert loop, inspired by Andrej Karpathy's autoresearch project.

    6.4K
  3. #3 · NousResearch/hermes-agent-self-evolution

    hermes-agent-self-evolution

    It is an evolutionary self-improvement system for Hermes Agent that uses DSPy and GEPA (Genetic-Pareto Prompt Evolution) to automatically optimize the agent's skills, tool descriptions, system prompts, and code, producing measurably better versions through reflective evolutionary search.

    5.5K
  1. 4

    EvoScientist

    5KEvoScientist/EvoScientist · Python

    EvoScientist is a self-evolving AI scientist system: a team of AI agents that autonomously explore research questions, generate insights, and improve themselves over time instead of waiting on constant human instructions.

  2. 5

    Zetta-Embodiment

    1.3Kair-embodied-brain/Zetta-Embodiment · Python

    Zetta is a closed-loop embodied harness that evolves code-based runtime critics and recovery skills online while keeping the base robot policy frozen, reaching 90.8% success on LIBERO-Pro and 93.6% on RoboCasa with an 11.1x inference speedup.

  3. 6

    Deepgraph

    1Kbillion-token-one-task/Deepgraph · Python

    DeepGraph is the autonomous research agent of JouleBeat, an open-source AI self-improvement (RSI) lab. It reads scientific literature, finds open questions, runs real experiments, and lets the results decide what is true.

  4. 7

    awesome-rsi

    766lobehub/awesome-rsi

    Awesome RSI is a curated research map of Recursive Self-Improvement (RSI): a taxonomy of papers, frameworks, benchmarks, and safety work on AI systems that improve their own capabilities and the mechanisms that generate further improvements.

  5. 8

    Motus2

    685shengshu-ai/Motus2

    Motus2 is a self-evolving general world model for dexterous robot manipulation. A single model with shared weights acts as a policy, a simulator, and an evaluator, coupling them into a closed decision-and-learning loop so the system can perceive, predict, act, and improve itself.

  6. 9

    wikiskill

    278ashutoshsinghpr7/wikiskill · Python

    WikiSkill is a faithful implementation of a Google Research paper (arXiv:2608.27454) that lets an AI agent build a persistent knowledge wiki from its own execution traces and use it to propose and validate changes to its own skill files, so lessons from failed runs survive instead of dying with the session.

  7. 10

    VibeGame

    272tettethu/VibeGame · Python

    VibeGame is an open-source, self-evolving multi-agent framework with an AI-native game engine that turns natural-language prompts into playable 2D web games, which you can then edit however you like. It runs on top of Claude Code and OpenAI Codex agent runtimes.

  8. 11

    dsh-evolve-modes

    204GraySilver/dsh-evolve-modes · TypeScript

    This is dsh-evolve-modes, an independent Web plugin for DeepSeek Harness (DSH) that adds a compact set of workflow controls — work state, thinking strategy, quality gate, and self-evolution — so you can compose how an agent works on each task, with self-improvement reviewed by a human before anything is applied.

  9. 12

    Harness-RSI

    170zgcagi/Harness-RSI · TeX

    Companion materials for our survey of how execution experience becomes persistent harness state under a fixed base model. The survey covers update mechanisms, evolutionary search, continual PAC learning, evaluation, applications, and societal impact.

  10. 13

    Proteus

    108proteus-evolve/Proteus · Python

    Proteus lets you plug any agent harness into any model, have it rewrite its own harness (prompts, memory, skills, tools, control loop) across many context-fresh episodes, and measures how that harness actually changes over time.

Often tagged together