primaxlab/selfevolvingai

SelfEvolvingAI is a Python framework described as a self-evolving AI system made up of 70 separate modules. The README is written primarily in Chinese. The project is marked as actively under development with APIs and functionality subject to change. The framework is organized around a loop: perceive inputs, recall from memory, reason about the problem, take action, reflect on what happened, and then update the system based on what was learned. Each stage in that loop is handled by dedicated modules. For example, memory is split into short-term, long-term, and episodic stores with forgetting curves. A self-improvement module analyzes its own code using Python's AST tools, detects issues, and attempts automatic fixes with version rollback available. A metacognition module evaluates confidence and detects gaps in the system's own knowledge. The 70 modules cover a wide range of AI-adjacent topics: knowledge graphs, causal reasoning, transfer learning, federated learning, reinforcement learning, prompt engineering, code generation, test automation, document generation, multimodal processing, defenses against adversarial attacks, vector storage, session management, a plugin system, stream processing, encryption, rate limiting, distributed locking, and internationalization, among others. The design principles listed in the README are: no external dependencies (pure Python only), each module works independently, all modifications must be reversible, changes are made in small increments rather than large rewrites, and all quality assessments use numeric scores. Running the system requires cloning the repository and calling main.py with one of several flags to run interactively, trigger an evolution cycle, view module status, or generate a report. The license is MIT.