alphaXiv has released OpenResearch, an open-source platform that turns everyday coding agents into full-fledged research assistants. The project hit GitHub Trending's number-one spot on September 17, crossing 1,000 stars in a single day.
OpenResearch gives Claude Code, Codex, OpenCode, and Cursor the ability to review literature, develop hypotheses, run experiments, and produce research artifacts -- essentially automating the entire scientific research loop. The platform supports what the team calls 'autoresearch,' where agents can propose an idea, change code, launch an experiment, inspect evidence, and decide what to try next, all without human intervention.
The platform is built around several core capabilities. Parallel exploration lets each research direction run in its own independent agent session with an isolated git worktree. A git-native experiment tree tracks every variant with immutable archives, making results reproducible. All logs, diffs, files, and artifacts stay tied to the work that produced them, keeping evidence in context.
OpenResearch runs entirely locally by default, storing data in SQLite on localhost. Projects, conversations, experiments, runs, logs, code, and artifacts all remain on the user's machine. An account is only needed for managed compute or organizational features. The same source snapshot can also run over SSH, on Slurm clusters, Kubernetes, Ray, Hugging Face Jobs, Modal, or Tinker.
The CLI (orx) integrates directly into supported agents through a skill installation. Common commands cover project management, run logs, experiment execution, literature discovery, and paper retrieval by arXiv ID or DOI. The platform supports local models through LM Studio, oMLX, Ollama, or custom endpoints.
