Choose the right project
A project groups research materials, datasets, jobs, saved results, and their history. Keep one investigation in one project so that you can trace how a result was produced. Switch projects from the workspace picker; check the selected project before uploading or starting a calculation.
Ask assigns work to your research group. Workflows connect agents and reusable tools into research processes you can run again.
Attach a CSV
A CSV upload creates a preserved research material and a dataset artifact. The original bytes remain unchanged. Derived audits, splits, and benchmarks become separate artifacts linked to that dataset.
Use UTF-8 CSV with unique, nonblank column names, consistent row widths, and at least three data rows. The supported intake allows up to 200 columns; row and upload-size limits depend on the server configuration.
- Choose the dataset file and inspect its columns.
- Tell your agents the actual citation, source URL, license, data type, and transformations you know.
- Leave unknown facts unknown. Use the synthetic-demo declaration only for the bundled demonstration file.
- Upload, then verify the material and dataset appear in the selected project.
Attach and extract a PDF
A PDF attachment preserves the original document. Text extraction is a separate job: accepting an upload does not mean extraction succeeded. Review its job status and extracted passages before relying on the paper as evidence.
Scanned papers may not provide usable text. OCR is not part of the documented ingestion workflow. An extraction failure does not remove the original attachment.
Understand access and provenance
An uploaded file is not automatically permission for every agent to read it. The workspace policy controls allowed projects, materials, artifacts, tools, and outbound access. Some single-operator installations explicitly enable automatic policy grants.
Provenance records the connections between inputs and derived results. Open History to answer “Which dataset, settings, and earlier steps produced this?” Source metadata is an assertion to verify, not proof that data is scientifically suitable.