You are here:
Document Reader Agent (Beta) in Agentforce Operations
Understand the capabilities, inputs, and limitations of the Document Reader agent (Beta) in Agentforce Operations.
Required Editions
| License Required |
|---|
This feature is a workspace that’s external to your Salesforce org and requires an Agentforce Operations license. To purchase an Agentforce Operations license, contact your Salesforce account executive. |
When to Use the Document Reader Agent
Use the Document Reader Agent to extract fields or tables from unstructured documents, such as PDFs or images, or you want to extract fields from documents where the format changes with each workflow run.
There are two different document agents, Learn more in Document Agents in Agentforce Operations
Capabilities
| Capability | Description | Example |
|---|---|---|
| Field value extraction | Extract values from structured or unstructured documents, such as PDFs and images. | Extract the vendor name, invoice date, and total amount from a scanned PDF invoice. |
| Find information with keywords | Provide different words to look for in a document that describe the information you want. Separate multiple aliases with commas. | For the PO Number field, enter: PO Number, PO #, Purchase Order #. |
| Instructions for individual fields | Add specific instructions about a field's format or location to help the agent find or interpret the value. You can also use instructions for conditional logic. | Tell the agent that Invoice Number is always a 6-digit number starting with 00, or add If missing, use today's date. |
| Supports fields with multiple values | For text fields, enter Yes in the Multiple Values column so the agent returns a comma-separated list of all matching values instead of a single value. | Extract all contact email addresses from a document that lists multiple contacts. |
| Tabular data extraction | Extract repeating rows of data such as line items. Request a table within a form in the Requested information section. Tabular data can’t be extracted directly into a CSV file. | Extract each line item from an invoice, including item description, quantity, unit price, and total. |
Configuring a Document Reader Agent Task
The Document Reader agent requires that you configure the task differently than other agents. Review the differences before configuring your task.
- In the Share information section, add the document field that
contains the file you want to extract data from. Pass the document to the task as a shared
field at run time.
Important Don’t upload files directly to the Documents input in the task setup. The agent only reads documents passed through shared fields. - In the Requested information section, add the fields you want the
agent to fill in:
- To extract system field values, add individual requested fields for each piece of data you need.
- To extract tabular data such as line items, add a form that includes a table. The agent fills in the table rows at run time.
Learn about how to Assign a Task to an AI Agent in Agentforce Operations
Field-Level Instructions
After you configure the requested fields on the task, a Provide Field Instructions for Agent section appears. Use this section to give the agent specific guidance on how to find and format each field's value. For general instructions that apply to all fields, such as Translate all content into English, use the task Description field instead.
For each requested field or table column, you can provide:
| Input | Description | Example |
|---|---|---|
| Keywords in Document | Different words to look for in the document that describe the information you want. Separate multiple aliases with commas. | For the PO Number field, enter: PO Number, PO #, Purchase Order #. |
| Additional Instructions | Other information about the data's format or location to help the agent find it. You can also provide conditional instructions. | A 6-digit number that starts with 00. Or: If missing, use today's date. |
| Multiple Values | Enter Yes if the field has more than one value. The value is returned as a comma-separated list of values instead of a single value. Available for text fields only. | If you're extracting contact email addresses and expect multiple emails, enter Yes. |
Limitations
- Supported file types: images, PDF, CSV, .docx, .xlsx, and .pptx.
- Pass documents to the task through a shared field. Uploading files directly to the Documents input isn't supported.
- Set up tabular data to be returned as a data grid configured through a form. The agent can’t extract tabular data directly into a CSV file.
- The agent can’t output Document, Signature, or Data Source fields.
- A maximum of 15 requested fields per task.
- Tables support a maximum of 10 columns and 50 rows.
