Skip to main content

Additional features: Upload file

When to use​

Enable Upload file when users need the agent to work with real documents or media—e.g., parse a PDF, summarize a DOCX, extract tables from an XLSX/CSV, register an image in a CMS, or hand a file to a downstream workflow or database.

Note: Uploading alone doesn’t give the LLM access to file contents. The agent must have a tool that can open the uploaded file via its link.

How it works​

  • When Upload file is ON, the chat composer shows an Upload option.
  • Each uploaded file is added to the user’s message as a link (file URL).
  • The agent can then pass that link to a tool (e.g., “Document Parser,” “Data Ingest,” “Image Classifier”) to read or process the file.
  • If no tool is available that accepts a file URL, the agent cannot use the upload in any meaningful way.

Enable in the UI​

  1. Open your agent → click **Edit.
  2. In Additional features, click Add.
  3. Toggle Upload file ON in the drawer (see screenshot).
  4. Close the drawer. You’ll now see Upload file listed under Additional features.

Prompting & tool setup (recommended)​

Add one or both of these to make uploads “just work”:

In the agent’s Base instruction​

Tell the agent which tool to call when a file is present and what to return.

“When the user uploads a PDF or DOCX, call DocumentExtractor with the file_url. Return a JSON summary with title, sections[], and key_values{}. If the file is an image, call ImageTagger with the file_url and return tags.”

In your tool(s)​

  • Ensure the tool schema includes a parameter like file_url (string).
  • Make sure the tool can fetch and process the file type(s) you expect (PDF, DOCX, XLSX/CSV, PNG/JPG, etc.).

Best practices​

  • Pair with the right tool: e.g., Parser for text, Ingestion for DB, Vision for images.
  • Be explicit: In prompts, specify expected output (JSON fields, table format, etc.).
  • Validate type/size in tools; instruct the agent to ask for a new upload if unsupported.
  • Chain actions: Parse first, then delegate to another tool/ sub-agent to store, enrich, or visualize.
  • Cite the source: Ask the agent to include the file name or link when summarizing results.

Troubleshooting​

  • Upload option not visible: Re-open Additional features → Add and ensure Upload file is toggled ON for this agent.
  • Agent ignores the file: You likely don’t have a tool that accepts a file_url, or the Base instruction doesn’t tell the agent when/how to use it.
  • Tool fails/timeout: Increase the tool timeout (Tools → per-tool gear), reduce file size, or handle errors in the tool to return a clear failure message.
  • Wrong tool chosen: Tighten Base instruction routing (e.g., “If extension is .pdf or .docx → DocumentExtractor; if .csv → TableImporter”).