momozi1996/momo-code

MOMO CODE — RSI *JEV*Pi --AI coding agent that evolves with you

What it solves

Momo Code is an AI-powered coding agent designed to evolve its performance over time. It addresses the limitation of static LLM coding assistants by implementing a self-evolution system that learns from successful patterns and user corrections to improve its coding tactics and decision-making.

How it works

The project utilizes a dual-speed evolution system based on the Pioneer Agent architecture:

  • Fast Loop (Experience): Uses the Knowledge Embedding Protocol (KEP) to distill successful session patterns into "Tactic cards." These are selected via Thompson sampling and injected into the system prompt in real-time to improve immediate task performance.
  • Slow Loop (Training): Performs hour-level weight improvements using Monte Carlo Graph Search (MCGS) and LoRA fine-tuning.
  • JEV Decision Layer: A Judgment-Evaluation-Verification layer that acts as a decision gate for tools, tactics, and routing, using a "shadow mode" to calibrate decisions against LLM outputs.
  • RSI Self-Evolution: A three-tier recursive improvement loop (Product, Strategy, Mechanism) that automates prompt patches and model weight updates.

Who it’s for

Developers who want a local-first, highly customizable AI coding agent that can be integrated into their CLI workflow and can autonomously improve its own strategies based on the project-specific context.

Highlights

  • Multi-Provider Support: Compatible with 25+ LLM providers including Deepseek, Claude, GPT-4, and Gemini.
  • Advanced Task Management: Supports recursive subagents for task decomposition and a Graph Engine for long-horizon tasks as resumable DAGs.
  • Holographic Capabilities: Includes a simulation agent for controlling physics worlds (Genesis) and a reasoning-driven optimization tool for parameter tuning.
  • Interactive REPL: A native TTY interface with a foldable Chain-of-Thought (CoT) panel and decision chips.
  • Claude Code Interop: Seamlessly inherits configurations and MCP servers from Claude Code.

Related

  • Project
    yologdev/yoyo-evolve

    yoyo is an open-source coding agent built in Rust that autonomously reads its own source code, decides what to improve, implements changes, runs tests, and commits—growing from 200 lines to over 158,000 lines in 180 days. It also provides a full REPL coding assistant with tools for editing, git, project health, and multi-provider support.

  • Project
    Zoo-Code-Org/Zoo-Code

    Zoo Code is a VS Code extension that adds an AI‑powered assistant capable of generating, refactoring, debugging, documenting, and automating code via natural‑language prompts. It offers several work modes, fast semantic search, safe command guarding, and integrates with many LLM providers.

  • Project
    divar-ir/ai-doc-gen

    An AI-powered documentation generator that uses a multi-agent architecture to analyze codebases and automatically create READMEs and AI assistant configuration files.

  • Project
    tobihagemann/turbo

    A collection of reusable Markdown-based workflows and skills for AI coding agents to standardize the process of planning, building, and shipping code.

  • Project
    CommandCodeAI/command-code

    An AI coding agent that builds software and continuously learns a developer's specific coding style and preferences through a taste profile.