Corollary Developers

Reference

Files

Structures, figures, tables, pages and meshes a run made, served as their own bytes. Every files[].url on a tool result points here.

GET/api/v1/files/{id}
bash
curl https://corollary-labs.com/api/v1/files/art_k3m9x2q7 \
  -H "Authorization: Bearer $COROLLARY_API_KEY" \
  -o structure.pdb

The response is the file itself — chemical/x-pdb for a structure, image/png for a figure, text/csv for a table, text/html for a page the agent wrote — with a Content-Disposition that names it. Ids begin art_.

What to expect

  • A file belongs to the account that made it. Another account's id is a 404, not a 403.
  • Files are kept for thirty days from when they were made, then 404. Run the tool again to regenerate one.
  • An HTML page is served as a document to save or parse, not to frame: it carries no script permissions here. Hosted, framable views are on the way — see the overview.
  • Links inside a tool's output and in the answer's Markdown already point at this endpoint, so a /api/v1/files/… URL found anywhere in a response can be fetched as is.