OpenResearch

OpenResearch is an MIT-licensed, open-source workspace for people using AI coding agents to conduct research. It gives each research direction a separate agent session and isolated Git worktree, so experiments can proceed in parallel without sharing the same working files. Desktop builds support macOS and Linux; the Windows beta requires Git for Windows. A local browser dashboard is also available.

It works with Claude Code, Codex, OpenCode, Cursor, and Google Antigravity. Choose an agent harness and model for each session. Local model configuration supports LM Studio, oMLX and Ollama; configured cloud agents and providers receive the requests sent to them. Agents can review literature, develop hypotheses, run experiments, and produce research artifacts. Its autonomous research loop can propose an idea, edit code, launch an experiment, inspect the results, and decide what to try next.

A Git-based experiment tree records how research branches relate to one another. Each run gets an immutable archive of its recorded commit, with logs, diffs, results, and artifacts linked to the work that produced them. This gives you a way to compare experiments and trace a result back to its code.

Compute can stay on your machine, run on your own infrastructure, or use remote services such as Modal and Hugging Face Jobs. OpenResearch also has a compute marketplace that brings together providers. Remote runs don't require publishing your repository.

Projects, conversations, code, and research records stay on your machine by default. An account is needed only for service-owned capabilities such as organizations and managed compute. Official builds send coarse usage analytics by default, with an opt-out. These events exclude code, prompts, file contents and paths.