A curated list of "autoresearch" systems and autonomous AI research agents, inspired by Karpathy's work. It targets researchers and engineers seeking to automate scientific discovery and software development via self-improving loops, offering a comprehensive overview of cutting-edge implementations and approaches.

How It Works

The core concept employs iterative "autonomous improvement loops" where AI agents propose, evaluate, and refine actions against a metric. This enables automated optimization of code, prompts, and research workflows. Novelty lies in generalizing this pattern across diverse domains and developing meta-agents that design or improve agent architectures, often outperforming manual methods.

Quick Start & Requirements

This is a curated list, not a single project. Users must consult individual project links for setup. Requirements vary but often include significant computational resources (GPUs, specific CUDA versions), large datasets, and LLM API keys. Setup complexity and resource footprint are project-dependent.

Highlighted Details

• Extensive coverage across general-purpose systems, research agents, platform ports (macOS, WebGPU), and domain-specific adaptations (genealogy, trading, GPU kernels).
• Numerous real-world use cases and detailed writeups, including optimizations for Shopify's Liquid engine and ancient ink detection.
• Links to relevant academic papers (e.g., ADAS, GEPA) and benchmark suites (MLAgentBench) for agent evaluation.
• Features projects on distributed/collaborative research and swarm intelligence.

Maintenance & Community

The list is actively curated, welcoming PRs and incorporating recent academic work. Specific community channels for the list itself are not detailed, but individual projects may provide them.

Licensing & Compatibility

The list is licensed under CC0-1.0 (public domain dedication), offering maximum reuse freedom. However, users must verify the licenses of individual projects linked within, as they may have different restrictions impacting commercial use or closed-source integration.

Limitations & Caveats

As a curated index, this repository requires users to evaluate each linked project individually. Many autoresearch systems demand substantial computational resources (high-end GPUs) and specific environments, posing adoption barriers. Projects are often research-oriented, potentially experimental, and may lack production-ready stability or comprehensive documentation.