alphaXiv/OpenResearch

What it solves

OpenResearch provides a local-first workspace designed specifically for research agents and autonomous research. It solves the problem of managing complex, iterative research workflows—such as literature review, hypothesis development, and experiment tracking—while maintaining local ownership of data and code.

How it works

The platform acts as a harness for existing coding agents (like Claude Code, Codex, or OpenCode). It allows users to run these agents in isolated git worktrees, enabling parallel exploration of different research directions. It uses a git-native experiment tree to track variants and ensure experiments are reproducible by creating immutable archives of recorded commits.

Who it’s for

It is designed for researchers and developers who want to automate the research loop (proposing ideas, modifying code, running experiments, and analyzing evidence) using AI agents while keeping their projects and logs on their own machines.

Highlights

• Autoresearch Loop: Capable of autonomously proposing ideas, changing code, and launching experiments.
• Local-First Architecture: Projects, conversations, and artifacts are stored locally using a SQLite store.
• Flexible Compute: Supports running locally, over SSH, or on various infrastructure like Slurm, Kubernetes, Ray, and Hugging Face Jobs.
• Agent Agnostic: Compatible with multiple agents, allowing the model and harness to be selected per session.
• Reproducible Experiments: Tracks experiment lineage and variants via a git-native tree.