Configuring a File agent

1. Navigate to your automation and create an agent responsible for getting the file (e.g. a Browser agent that will download it from a webpage, etc.) and a File agent responsible for extracting information from the file

2. Instruct the first agent on how to locate the file, then instruct it to call the File agent

3. Configure the File agent to expect the file(s) as input(s)

4. Simply name the file appropriately and give it an optional description, then click Save

5. Create the data points that you wish to extract from the file(s) as outputs

6. Simply name the output appropriately and in the Description field, prompt the agent on how to locate the data you're looking for, then click Save

7. Remember to click "Save" to save your progress regularly

TLDR: If it’s a document humans read visually, convert it to PDF first. If it’s structured machine data, prefer JSON or CSV. If it’s an Office file, treat it as unsupported unless you add a conversion step upstream.

Use PDF, PNG, JPG/JPEG, WEBP, common audio files, and common video files. These are the formats most likely to work well with the default extract-agent path, especially PDF for documents and scanned forms.

Use JSON, TXT, CSV, and other plain-text files only when the content is mostly text and layout does not matter. They can work, but extraction quality depends on the text being clean and self-explanatory.

Avoid DOCX, XLSX, PPTX, DOC, XLS, and similar Office formats. Magical won’t block them at upload time, but the default extract path does not convert them into a format the model is documented to understand well.