This repository implements a self-evolving AI agent with 36 tools organized into 7 domain modules. The system uses a three-layer memory architecture combining session history, LLM-compressed long-term memory, and LanceDB vector retrieval. It supports MCP servers via JSON-RPC, dynamic tool filtering across 5 context profiles, a nudge system for behavioral correction, and a circuit breaker mechanism for tool failure handling. The agent runs function-calling loops with up to 20 iterations, auto-creates runtime tools, and operates continuously in production environments.