Overview
OpenAI4S, from PKU-YuanGroup, coordinates scientific tasks through structured tools and persistent analysis kernels. Python cells can call host services during execution; R is a separate analysis channel. The workbench retains execution records and versioned outputs so users can inspect the steps behind a result.
Chemistry work can combine PubChem or ChEMBL retrieval with optional RDKit analysis and searchable workflow recipes. These are documented software workflows; generated conclusions still require scientific checking. The package declares Alpha status. Remote computing and model-endpoint integration have implementation limits, and physical laboratory control is future work. The project independently draws on Claude Science concepts; it is not presented as an official OpenAI or Anthropic product.
Key Features
- Persistent Python and R analysis channels with structured JSON orchestration; mid-cell host calls are Python-only.
- PubChem and ChEMBL connectors return structured chemical records; PubChem uses exact names or synonyms rather than fuzzy search.
- Optional science dependencies provide dataframe and plotting tools; the chemistry extra adds RDKit where wheels are available.
- Searchable code-based Skills include chemistry and literature workflows; recipes do not provision all optional software.
- External MCP connector tools can list servers, discover tools and call them. This establishes a client integration, not a standalone MCP server.
- The browser workbench exposes execution traces, versioned artifacts and provenance; the Notebook is read-only by default.
Use Cases
- Retrieve compound identifiers and properties, then inspect or analyze them with an available RDKit environment.
- Combine chemical records with literature metadata to prepare a source-attributed research report.
- Review Python/R data-analysis outputs and their execution history before reusing figures or tables.
How to Use
Requires Python 3.10+ and a supported operating system. In a dedicated Python environment, the documented PyPI route can be expressed as:
python -m pip install "openai4s[science,chemistry]==0.3.1"
openai4s serve
These are documentation examples only. In Settings → Models configure your own supported provider and select an active model. Configure search separately if needed. Add R through the documented Conda-family setup; R is not bundled. Request a small compound lookup first and inspect the returned source and output.
Native Windows kernels are unsupported; use the documented WSL2 route. Version 0.3.1 has no macOS DMG. The gateway API has documented gaps and no OpenAPI specification; do not assume complete API coverage. Read the release and upgrade instructions before changing an existing installation.
Sources and verification
This entry describes capabilities documented by the linked project sources. Catalog edits are not evidence of a successful run, compatibility test or scientific performance evaluation.
Read the editorial and verification policy