momozi1996/momo-code

MOMO CODE — AI coding agent that evolves with you

What it solves

momo Code is an AI-powered coding agent designed to evolve alongside the user. It addresses the limitations of static AI assistants by implementing a dual-speed self-evolution system that allows the agent to learn from its successes and failures in real-time and over longer periods, improving its coding tactics and model weights automatically.

How it works

The system uses a two-tier evolution architecture:

  1. Fast Loop (/evolve): Uses the Knowledge Embedding Protocol (KEP) to distill successful patterns into "Tactic cards." These are selected via Thompson sampling and injected into the system prompt in seconds to immediately improve performance.
  2. Slow Loop (/fine-tune): Performs hour-level weight improvements using Monte Carlo Graph Search (MCGS) and LoRA fine-tuning to permanently bake learned behaviors into the model.

Additionally, it features a recursive subagent system for task decomposition, a graph engine for managing long-horizon tasks as resumable DAGs, and a simulation agent that can control a Genesis physics world via Python code.

Who it’s for

Software developers who want an autonomous coding agent that can handle complex, long-running tasks, integrate with various LLM providers, and improve its own performance based on the specific patterns of the user's codebase.

Highlights

  • Dual-Speed Evolution: Combines instant prompt-based tactic injection with long-term LoRA fine-tuning.
  • Multi-Agent Orchestration: Supports recursive subagents and DAG-based graph execution for complex workflows.
  • Extensive Provider Support: Compatible with 25+ LLM providers and any OpenAI-compatible API.
  • Simulation Integration: Includes a /sim agent for controlling physics worlds and a /optim tool for reasoning-driven parameter tuning.
  • Local-first & Interoperable: Keeps code on the local machine and allows seamless migration from Claude Code.

Related

  • Project
    Kilo-Org/kilocode

    An open-source AI coding agent for VS Code, JetBrains, and the CLI that provides access to 500+ models for code generation, debugging, and autonomous CI/CD operations.

  • 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
    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.

  • Project
    Cranot/roam-code

    roam‑code is a local static‑analysis CLI (and optional MCP server) that builds a symbol graph of a repository, letting AI coding agents query definitions, callers, test impact, and risk (blast radius) without any remote API calls. It provides pre‑flight checks, health summaries, verification gates, and Claude‑specific hooks, all open‑source and privacy‑preserving.

  • Project
    evo-hq/evo

    An autoresearch orchestrator that automates code optimization by running parallel AI-driven experiments in a loop to iteratively improve codebase metrics.