Turn your coding agent into a research partner with a local-first workspace
Imagine a coding tool that handles everything from paper retrieval to experiment design and result logging on its own. OpenResearch extends existing coding agents for research use. With this tool, researchers can automate literature reviews and experiment iterations without writing code themselves.
Extending coding agents for research
OpenResearch leverages coding agents you already use, such as Claude Code, Codex, and Cursor. It supports workflows where these agents review the latest literature, formulate hypotheses, and run experiments. It integrates with alphaXiv, bioRxiv, and PubMed to find relevant papers and connect them with existing research and experiments. Experiments can be run locally or executed in various compute environments, including SSH, Slurm, Kubernetes, and Modal.
Experiment logging and data sovereignty
All experiments are recorded in a local SQL database, with logs, code, and artifacts remaining on the user's machine. The agent references this historical record to determine the direction of the next experiment. The desktop app allows you to easily verify what the agent did and the basis for its results. Coding agents, prompts, file contents, repository names, tokens, emails, and project identifiers are not collected. All data remains under the user's control.
Installation and initial setup
It is available on macOS 11 and later, Windows (beta, requires Git for Windows), and Linux (glibc 2.35 or later). You can install it by downloading the desktop app or using the curl -LsSf https://openresearch.sh/install.sh | sh command. In managed Mac environments, the CLI may be blocked because it is unsigned, so the signed desktop app is recommended. To use the orx command in the terminal, you can install it from the Updates tab in the app settings.
Pre-use considerations
Official release builds send coarse product analytics events and agent feedback linked to a random installation ID in the background. This data is used to understand model usage, report bugs, and handle feature requests, while excluding sensitive research information or code. You can completely disable this feature using the orx telemetry off command or through the app settings. When using local models, you can integrate with OpenCode via LM Studio, oMLX, Ollama, or a custom endpoint. Creating an account on openresearch.sh allows you to use email updates and managed compute services.
