Skip to main content
Ask Erdo to download files from a link and it uses its download_files tool to keep them on a Knowledge record: logos, fonts, brochures and other files from a direct file URL or a public Dropbox, Google Drive, SharePoint or OneDrive link. Anyone-with-the-link shares work without connecting an account. Files can be up to 50 MiB each. Create the destination record first, then name it as record. You need edit access to that record in your selected organization. Downloaded attachments stay private until you explicitly publish or share them. Use POST /v1/shared-links/inspect, the MCP tool erdo_inspect_shared_link, or erdo files inspect-link <url> to check a public file or shared-folder link. Inspection does not import files, create records, or store content. Keep all sharing parameters in the URL. Only HTTPS URLs without credentials are accepted.
The REST request is:
An accessible file returns:
provider is sharepoint, onedrive, dropbox, google_drive, or web. kind is file, folder, or unknown. opens: true means file content or a structured folder listing was reached without signing in. It does not guarantee that a later import succeeds, or that every file in a folder is downloadable. When opens is false, reason explains the result:
  • sign_in_required: an authentication endpoint, sign-in form, or HTTP 401/403 was returned. Ask the owner for an anyone-with-the-link share.
  • not_found: the server returned HTTP 404 or 410.
  • unsupported: access could not be established from the response. This is inconclusive, including HTML previews that need a browser to inspect.
  • error: the request failed, timed out, or was blocked by public-URL checks.
Inspection has a 20-second budget, reads only bounded content prefixes, and returns at most 50 sample files. SharePoint folders use the same guest session and REST listing as imports, with limits on response size, folder depth and folder count. file_count is null when enumeration is incomplete; total_bytes is null when enumeration or sizes are incomplete. Unknown sample sizes are null. complete is false whenever enumeration is truncated or a size is unknown. Inspection makes no hosted-browser calls and incurs no paid browser lookup. Dropbox, Google Drive and personal OneDrive folder previews may therefore return unsupported, even when their owner has shared them publicly. Inspection requires an authenticated member of the selected organization and supports API keys.

Run the download

The agent calls download_files with:
From the CLI or MCP, send the request to an agent (for example erdo agent send) and it calls download_files the same way. Keep all sharing parameters in the original link. SharePoint and OneDrive for Business links are read directly with the guest access the link grants. Google Drive file links also use direct downloads. Dropbox, Google Drive and personal OneDrive folders download inside a hosted browser session. The download never signs in or changes sharing permissions. A link that requires a login fails with denied; its owner must make it accessible, or replace it with a public file link. For a single file, Knowledge attachment upload accepts these public file share links through the CLI, REST and MCP as well. Folder URLs sent to attachment upload return an explanation directing you to this importer.

Select files and continue

Omit paths to download everything. To select specific material, supply exact relative filenames or folder prefixes ending in /, such as ["logos/", "brochures/facts.pdf"]. Nested folders are included by default; set recursive to false to select root files only. One call processes up to ten files. If listing_complete is false and next_cursor is present, pass it as cursor with the same link, record and filters. If there is no cursor, check the reported failure and retry the original request. Folder contents can change between calls; start a fresh import to include files added before an earlier page’s cursor. listing_complete describes the folder listing, including files that failed to download. Check failures even when it is true. Each failure has a filename, source URL and reason: denied, too_large, unsupported, timeout or unreadable. Retry failed paths separately after resolving the cause. Exact replays of the same filename and bytes reuse the existing attachment.

Read the result

Each imported file includes its filename, media type, byte size, original source URL, attachment_id and attachment_handle. A PDF also has an extraction_status:
  • extracted: text_handle is a stored-content handle. An agent reads it with read_document(dataset_slug=text_handle) or searches it with read_document with a query.
  • unreadable: the PDF is stored, but readable text was not obtained. This can happen with scanned, encrypted or malformed PDFs, or if text storage failed.
  • not_applicable: the imported file is not a PDF.
The original PDF appears when you read the Knowledge record. Fonts remain binary attachments; importing a font does not make it available for use in a published page.