Skip to main content

Files and Document Extraction

Fased has a built-in Files / document extraction service. It is configured from Agent > Services and uses the gateway file limits under gateway.http.endpoints.responses.files. This is the Fased replacement for copying a generic document-extract extension: document handling is a Media/Files service contract, not a random extension card and not an Agent model provider.

What It Does

The service extracts bounded text from attachments before the Agent sees them:
  • local chat/task/channel attachments;
  • OpenResponses input_file parts;
  • URL-based file inputs when URL fetch is allowed;
  • text, Markdown, HTML, CSV, JSON, and PDF.
PDFs use text extraction first. If a PDF has too little embedded text and the optional canvas dependency is available, Fased can render bounded page images for model-side understanding.

Configure

Open Agent > Services > Files / document extraction. Common fields:
  • Allow URL files: lets OpenResponses file inputs fetch remote documents.
  • Allowed MIME types: restricts file types. Leave empty for the built-in conservative defaults.
  • URL allowlist: restricts remote file hosts. Use exact hosts or *.example.com.
  • Max bytes / max chars: caps how much data is fetched and injected.
  • PDF pages / pixels / text threshold: caps PDF parsing and fallback render.
Example raw config:

Access Control

The Files service configures extraction. It does not grant every Agent file tools automatically.
  • Use Agent > Tools to allow or deny file/media tools per Agent.
  • Keep URL fetching disabled or allowlisted when the gateway is reachable from a hosted environment.
  • Treat extracted document text as untrusted external content.