Autoresearch — AI Agent Framework: Live Stats & TrendScore

Live GitHub stats, community sentiment, and trend data for Autoresearch. TrendingBots tracks star velocity, fork activity, and what developers are saying — updated from real data sources.

GitHub data synced: May 2, 2026 • Sentiment updated: Apr 18, 2026

GitHub Statistics

Community Sentiment

Community Buzz: Andrej Karpathy's 'autoresearch' is 'an autonomous loop where AI edits PyTorch, runs 5-min training experiments, and continuously lowers its own val_bpb' as quoted on Reddit, and 'Karpathy's Autoresearch Lets AI Agents Run ML Experiments Alone' on X/Twitter

Pros & Cons

What People Love

Reddit users praise autoresearch for its potential to accelerate AI research, Dev.to users appreciate the innovation in AI agents running ML experiments

Common Complaints

Limited local execution capabilities, AI models 'lying and cheating'

Biggest Positive: AI research innovation

Biggest Negative: Limited local execution

Why Autoresearch Stands Out

Autoresearch is different from alternatives because it provides a simple, yet powerful way to create autonomous agents that can improve themselves over time. By following the principles of one metric, constrained scope, fast verification, and automatic rollback, Autoresearch enables users to create agents that can learn and adapt without manual intervention. The project's use of git as memory also allows for the preservation of failed experiments, enabling the agent to learn from its mistakes. Additionally, Autoresearch's ability to generalize to any domain makes it a valuable tool for a wide range of applications.

Built With

Build an autonomous research assistant that iterates on a goal-directed task — Autoresearch enables this by providing a loop that never quits, with mechanical verification and automatic rollback, Build a relentless improvement engine for Claude Code, OpenCode, or OpenAI Codex — Autoresearch turns these into autonomous agents that modify, verify, and keep or discard changes, Build a system that generalizes Karpathy's autoresearch principles to any domain — Autoresearch makes this possible by following simple principles: one metric, constrained scope, fast verification, and automatic rollback, Build a project that uses git as memory to preserve failed experiments in history — Autoresearch allows the agent to read git log and git diff to inform its decisions, Build a loop that performs mechanical verification and logs results in TSV format — Autoresearch provides this functionality to enable progress tracking and analysis

Getting Started

  1. Install Autoresearch by running the command `git clone https://github.com/uditgoenka/autoresearch.git`
  2. Configure the agent by setting the goal, scope, and metric in the `config.json` file
  3. Run the agent using the command `./autoresearch loop`
  4. Verify that the agent is working by checking the `results.log` file
  5. Try running the agent with a specific scenario, such as `./autoresearch scenario --name my_scenario`, to verify that it works

About

Claude Autoresearch Skill — Autonomous goal-directed iteration for Claude Code. Inspired by Karpathy's autoresearch. Modify → Verify → Keep/Discard → Repeat forever.

Official site: https://udit.co/projects/autoresearch

Category & Tags

Category: research

Tags: ai, autonomous-agent, autoresearch, claude, claude-code, iteration, karpathy, productivity, skill

Market Context

Autoresearch is seen as a competitive edge in the AI research space, with Andrej Karpathy's work being widely discussed