AlphaXiv published OpenResearch, an MIT-licensed, local-first workspace that turns Claude Code, Codex, OpenCode, or Cursor into research agents that review literature, develop hypotheses, run experiments, and produce research artifacts. The orx CLI launches a local dashboard, and runs can target local machines, SSH, Slurm, Kubernetes, Ray, Modal, or managed compute.
OpenResearch treats coding agents as research workers by giving each direction its own session, git worktree, and recorded experiment lineage. Variants are tracked in a git-native experiment tree, and every run receives an immutable archive of its recorded commit, so logs, diffs, and results stay tied to the code that produced them. In autoresearch mode the tool can propose an idea, change the code, launch an experiment, inspect the evidence, and decide what to try next.
Coding agents such as Claude Code, Codex, OpenCode, and Cursor are now capable enough that teams want to point them at experiments rather than only product code. Anchoring agent state in commits and artifacts rather than chat history makes intermediate research steps inspectable by humans. Projects, runs, and logs stay on the local machine in a SQLite store by default.
The repository describes capabilities but provides no evidence that autonomously produced research results are correct or reproducible in practice. The README also warns that the remote mode binds to loopback without application-level authentication, so other users on that host can reach it. A meaningful test will be published case studies where OpenResearch runs are independently reproduced from their archived commits.