yologdev/yoyo-evolve

A coding agent that evolves its own source, in public — 200 lines of Rust on day one, every commit since agent-written and tests-gated.

What it solves

yoyo is a self-evolving coding agent designed to automate the software development lifecycle. It eliminates the need for human-written code by autonomously reading its own source, planning improvements, implementing changes, running tests, and committing updates to its own repository. It also handles common developer tasks like codebase navigation, multi-file edits, and git management within a terminal-based REPL.

How it works

yoyo operates through a series of autonomous cycles:

  • Evolution Cycles: Every 3 hours, the agent reads its own source code and GitHub issues, plans improvements, implements them, and verifies them with tests before committing changes.
  • Social Sessions: Every 4 hours, it interacts with GitHub Discussions to learn from humans and contribute to conversations.
  • Memory Synthesis: Daily, it compresses and regenerates its active memory from JSONL archives to maintain a persistent, time-weighted context for its prompts.
  • Agent Core: It uses a streaming REPL with support for multiple LLM providers, sub-agent spawning for parallel task execution, and an automatic retry system with exponential backoff.

Who it’s for

Developers who want a powerful, autonomous coding assistant in their terminal that can manage complex projects, perform structural refactoring, and handle the project's own maintenance.

Highlights

  • Self-Evolution: Grows its own codebase autonomously from a small initial seed of Rust code.
  • Extensive Toolset: Includes a comprehensive suite of slash commands for git integration, project health checks, structural code search (via ast-grep), and risk scoring.
  • Multi-Provider Support: Compatible with 15 different LLM providers including Anthropic, OpenAI, and Google.
  • Advanced Memory: Features project-specific context files, persistent memories, and a time-weighted synthesis job for long-term learning.
  • Security-First: Implements a robust permission system with directory restrictions and interactive tool approval to prevent unauthorized executions.

Related

  • Project
    ANative-Lab/EvoAgentX

    An open-source framework for automatically building, evaluating, and evolving LLM-based agentic workflows through iterative feedback loops.

  • Project
    PrimeIntellect-ai/prime-agent

    Prime Agent is an open‑source AI coding/research assistant that runs a persistent Python REPL powered by a Recursive Language Model (RLM). It stores supplemental prompts, memories, and reusable skills in a durable “Continual Harness” that can be refined incrementally. The tool supports background, daemon‑backed sessions, sub‑agent spawning, direct agent‑to‑agent communication, scheduled heartbeats, persistent goals, and a bounded autonomous mode. Install via a verified shell script, authenticate a provider, and run `prime-agent` in any project directory. Licensed under MIT and accompanied by extensive documentation and an arXiv citation.

  • Project
    github/app

    The GitHub Copilot app is a desktop client that lets you run Copilot‑powered AI agents in isolated local or cloud workspaces, manage multiple parallel sessions, and oversee the entire coding workflow—from picking a task to merging a PR—through visual canvases and automation tools.

  • Project
    smallnest/pigo

    pigo is a Go‑based CLI AI coding assistant (a clone of the pi agent) that can read/write files, run shell commands, fetch web content, and interact with many LLM providers. It works in headless script mode or an interactive REPL, supports tool‑level allow/deny lists, project trust, prompt templates, skills, plugins, self‑update, and provides a Go SDK for embedding.

  • Project
    ColeMurray/background-agents

    An open-source background coding agent system that executes software engineering tasks in isolated sandboxes and submits results as Pull Requests.