Awesome RSI

Papers in which the loop that improves a system also changes the thing doing the improving.

🧬 Rewrites its own code · ♻️ Optimizes its own optimizer · 🧩 Edits the harness that edits · 🌱 Searches its own designs · ♾️ Formal machines

• 🔥 News
• 🧬 Rewrites Its Own Code (10)
• ♻️ Optimizes Its Own Optimizer (12)
• 🧩 Edits the Harness That Edits (4)
• 🌱 Searches Over Its Own Designs (4)
• ♾️ Formal Self-Referential Machines (9)
• 🔗 Where the Rest of the Field Is (10)

🚀 2026-09 · Repository launch. PRs welcome.

The agent's working directory contains the agent, so an edit can land on the part that decides what to edit next.

• ⭐ Darwin Godel Machine, "Open-Ended Evolution of Self-Improving Agents".
• ⭐ Gödel Agent, "A Self-Referential Agent Framework for Recursive Self-Improvement".
• ⭐ A Self-Improving Coding Agent.
• Hyperagents.
• Group-Evolving Agents, "Open-Ended Self-Improvement via Experience Sharing".
• Huxley-Gödel Machine, "Human-Level Coding Agent Development by an Approximation of the Optimal Self-Improving Machine".
• Live-SWE-agent, "Can Software Engineering Agents Self-Evolve on the Fly?".
• MOSS, "Self-Evolution through Source-Level Rewriting in Autonomous Agent Systems".
• Bounded Recursive Self-Improvement.
• Mendel Gödel Machine, "Recursive Self-Improving Coding Agents via Comparative Evolution".

The improvement is produced by a procedure, and that procedure is applied to itself.

• ⭐ Self-Taught Optimizer (STOP), "Recursively Self-Improving Code Generation".
• ⭐ Self-Adapting Language Models.
• ⭐ Promptbreeder, "Self-Referential Self-Improvement Via Prompt Evolution".
• Meta^n, "Recursive Self-Improvement through Emergent Depth".
• metaTextGrad, "Automatically optimizing language model optimizers".
• SePO, "Self-Evolving Prompt Agent for System Prompt Optimization".
• MetaSkill-Evolve, "Recursive Self-Improvement of LLM Agents via Two-Timescale Meta-Skill Evolution".
• Learning to Evolve, "A Self-Improving Framework for Multi-Agent Systems via Textual Parameter Graph Optimization".
• Metalearning Continual Learning Algorithms.
• Arbitrary Order Meta-Learning with Simple Population-Based Evolution.
• Eliminating Meta Optimization Through Self-Referential Meta Learning.
• A Modern Self-Referential Weight Matrix That Learns to Modify Itself.

The scaffold around the model is the substrate, and the component doing the scaffold editing is inside the substrate.

• Ouroboros, "A Self-Developing Frontier Coding Agent with Reviewed Core Evolution".
• Continual Harness, "Online Adaptation for Self-Improving Foundation Agents".
• Self-Harness, "Harnesses That Improve Themselves".
• EvoTrainer, "Co-Evolving LLM Policies and Training Harnesses for Autonomous Agentic Reinforcement Learning".

A meta level writes candidate agents, operators or memory architectures as code, then searches the archive it just extended.

• ⭐ Automated Design of Agentic Systems.
• The Red Queen Gödel Machine, "Co-Evolving Agents and Their Evaluators".
• Learning to Continually Learn via Meta-learning Agentic Memory Designs.
• AlgoEvolve, "LLM-driven Meta-evolution of Algorithmic Trading Programs".

Machines and theorems, where the question was first posed: what self-modification can be proved to buy, and at what cost.

• ⭐ Godel Machines, "Self-Referential Universal Problem Solvers Making Provably Optimal Self-Improvements".
• From Seed AI to Technological Singularity via Recursively Self-Improving Software.
• Self-Reference in Large Language Models: The Introspection Threshold for Recursive Self-Improvement.
• Performance of Bounded-Rational Agents With the Ability to Self-Modify.
• A Formulation of Recursive Self-Improvement and Its Possible Efficiency.
• Self-Modification of Policy and Utility Function in Rational Agents.
• The Unverifiability of Artificial General Intelligence (AGI) Alignment, Static and Dynamic: From Trakhtenbrot's Wall to the Safety-Generality Tension.
• What does a system modify when it modifies itself?.
• SGM, "A Statistical Godel Machine for Risk-Controlled Recursive Self-Modification".

Self-improving and self-evolving agents, prompt optimization, memory and skill libraries, benchmarks and safety work are all out of scope here. These lists cover them.

• Awesome AI Scientist, Sibling list, for AI systems that do science rather than improve themselves.
• awesome-rsi (lobehub), Research map of RSI organized by model level, harness level and automated AI R&D.
• awesome-rsi (pinkbubblebubble), Evidence-labelled collection with an explicit inclusion decision procedure.
• Awesome-Self-Evolving-Agents, Companion list to the what, when, how and where to evolve survey.
• Awesome-Self-Improving-Agents, Reading list for self-improvement in foundation-model agentic systems.
• Awesome-Harness-Self-Improvement, Focused on the harness layer, bilingual, with an explicit optimization ladder.
• awesome-recursive-self-improving-agents, Living index for the foundation, framework and future directions survey.
• Awesome-Agent-Harness, Survey-backed list of agent harness designs.
• awesome-automl-papers, The pre-LLM automated machine learning literature this field grew out of.
• Awesome-RL-for-LRMs, Reinforcement learning for large reasoning models, the training half of the weight-level loop.

Open a pull request. Link the paper, add the code repository if there is one, and say in one line which component of the system the improvement loop modifies. See CONTRIBUTING.md for the entry format.

@misc{awesome_rsi,
  title        = {Awesome RSI},
  year         = {2026},
  howpublished = {\url{https://github.com/Omni-Scientist/Awesome-RSI}},
  note         = {Papers in which the improvement loop modifies its own machinery}
}