Reads or shares workspace files. Three actions are available: 'read' — return file contents inline. For PDFs: (1) If the user specifies a page number, read that range directly using pageStart/pageEnd. (2) If searching for a topic, first read pages 1-10 to find the Table of Contents. If no TOC is found in pages 1-10, extend to pages 1-20, then check the final 10 pages as some PDFs place the index at the back. (3) If the PDF has no TOC, scan in 15-page chunks from the beginning until the topic is located. (4) Always read in chunks of 20 pages or fewer. Never request the full PDF in a single call. (5) If a section appears to continue beyond the chunk boundary, read the next chunk to complete it. (6) For PDFs under 15 pages total, reading the entire document in one call is acceptable. 'preview' — REQUIRED first step before sharing. Returns file metadata (name, size, type, modified) and a short-lived confirmationToken. Present ALL metadata to the user and ask for explicit approval before proceeding. Never skip this step. 'download' — generate a public download link. Requires the confirmationToken returned by a prior 'preview' call for the same file. A link CANNOT be generated without a valid token.

Downloads a file directly from a public or authenticated HTTP/HTTPS URL and saves it to the workspace.

Searches the live internet for accurate, up-to-date information. Use for current events.

Saves a file provided directly by the AI agent into the workspace. Use this to save generated images, PDFs, DOCX, or text files without needing an external URL download. Binary files MUST be provided as base64 encoded strings.