karpathy/autoresearch

This page introduces an external open-source repository. It is not an HDATF product.

What it does

A repository in which an AI agent runs small LLM training experiments on its own, editing code, training for five minutes, and keeping or discarding each change based on the result. People guide it through a program.md file.

How it helps ATF

A reference for LabChin and Harness on an experiment loop that tries a change, checks the result, keeps or discards it and logs each run. Writing research direction in a Markdown file for the agent is also worth comparing.

License

No license No license file found. Without a license, reuse of the code is not permitted by default.

More in this category

• AstrBotDevs/AstrBot
  An open-source agent chatbot platform that connects LLMs to messaging apps such as QQ, WeChat Work, Feishu, DingTalk, Telegram and Slack. It supports MCP, skills, a knowledge base, plugins and an agent sandbox for running code.
• paperclipai/paperclip
  A Node.js server and React UI for running teams of AI agents toward business goals. Agents such as Claude Code, Codex or Cursor get goals and budgets, and people approve strategy and track work and costs from one dashboard.
• NousResearch/hermes-agent
  Hermes Agent is a self-improving AI agent from Nous Research. It creates skills from experience and improves them in use, keeps agent-curated memory, searches past sessions, and can be reached from Telegram, Discord, Slack or a terminal.
• agentscope-ai/QwenPaw
  QwenPaw is a personal AI assistant that runs locally or in the cloud, connects to chat apps and is extended with skills and plugins. It keeps three layers of memory, from working context to a linked Markdown knowledge base.
• alibaba/page-agent
  A JavaScript library that adds a GUI agent inside a web page so users can control the interface with natural language. It works on the DOM as text without screenshots, and offers an optional Chrome extension and MCP server.

Only repositories in the ranked Trendshift lists are included, and the lists are used only to find candidates. We do not copy their ranks. Descriptions, licenses and star counts come from each GitHub repository. The notes are our own reading. We have not tested these projects, and a place on a trending list does not prove quality.