Overview

OpenResearch empowers researchers by transforming AI models like Claude Code, Codex, and OpenCode into sophisticated research agents. This local-first workspace enables parallel exploration of research directions, reproducible experimentation, and seamless integration of evidence within context. Built for autoresearch, it can autonomously manage the entire research loop, from hypothesis generation to artifact production.

Key Features

• Parallel Agent Exploration: Assign independent agent sessions to distinct research directions, each with its own isolated git worktree for clean separation.
• Reproducible Experimentation: Leverage a git-native experiment tree to meticulously track variants and ensure every run is archived with its immutable commit.
• Contextualized Evidence: Maintain a tight link between logs, code diffs, files, experimental results, and the artifacts they produced.
• Flexible Agent & Compute Choice: Utilize Claude Code, Codex, or OpenCode, selecting the harness and model per session. Run computations locally, on your own infrastructure, or via managed OpenResearch compute.
• Local Ownership & Privacy: All projects, conversations, experiments, runs, logs, code, and artifacts are stored locally on your machine by default, with no code or sensitive data published without explicit action.
• Autoresearch Capabilities: Automate the full research lifecycle, from proposing ideas and modifying code to launching experiments and iteratively refining next steps.
• Run Anywhere: Deploy the workspace locally, over SSH, or on various compute platforms like Slurm, Kubernetes, Ray, Hugging Face Jobs, Modal, and Tinker.
• Seamless CLI Integration: Interact with OpenResearch through a powerful command-line interface, enabling project management, run tracking, log inspection, and discovery of research papers.

Typical Use Cases

• AI-driven Literature Review: Agents can autonomously scan, summarize, and synthesize information from research papers to identify knowledge gaps or novel hypotheses.
• Automated Hypothesis Testing: Develop and test scientific hypotheses by having agents generate experimental designs, write simulation code, and analyze results.
• Code Refinement and Optimization: Agents can explore different code modifications for performance or accuracy improvements, with each variation tracked as a distinct experiment.
• Multi-Agent Collaboration on Complex Problems: Deploy multiple agents in parallel to tackle different facets of a research problem, with the experiment tree preserving the lineage of each contribution.
• Reproducible Scientific Discovery: Ensure that all aspects of a research project, from initial setup to final results, are version-controlled and easily reproducible.
• Remote Research Execution: Run computationally intensive research tasks on powerful remote servers while managing the entire process from your local machine.