Extract tables from files

Description

The Extract tables from files activity sends input files to the Infoveave AI service and extracts tabular information from them. It can process all files together or one file at a time. It can also guide extraction using a column alias map or a custom structured response schema.

Supported Features

File-based table extraction: Extract tables from one or more attached files.
One file at a time mode: Process each file independently and add a FileName column to extracted rows.
Combined file mode: Send all files in a single AI request.
Column alias map: Define expected output columns and aliases that may appear in source documents.
Generated schema: When structured output is disabled, the activity builds a response schema from the configured column map.
Custom structured output: When enabled, provide a custom JSON response schema.
Raw response retention: The raw AI response is returned in AdditionalResponse.

Input

Type	Required	Description
Files	Optional	Files passed to the AI service as inline attachments. Each file is converted to base64 and sent with a detected MIME type.

Input Scenarios

1. No Files

The activity can run without files, but extraction normally requires attachments. If the AI service does not return rows, the activity fails.

{
  "Files": []
}

2. Single File

{
  "Files": [
    { "FileName": "invoice.pdf", "FullPath": "C:/Work/invoice.pdf" }
  ]
}

3. Multiple Files

{
  "Files": [
    { "FileName": "invoice-001.pdf", "FullPath": "C:/Work/invoice-001.pdf" },
    { "FileName": "invoice-002.pdf", "FullPath": "C:/Work/invoice-002.pdf" }
  ]
}

Output

The activity reads the Response property from the AI JSON response and converts each array item into one data row.

Field	Type	Description
Data	Array	Extracted table rows. Column names depend on `Column Map` or the custom `ResponseSchema`.
Errors	Array	Parsing errors or execution errors, if any.
AdditionalResponse	String	Raw AI response JSON. In one-file-at-a-time mode, this currently contains only the combined response variable and may be empty even when per-file responses succeeded.

Example Output

{
  "Data": [
    {
      "InvoiceNumber": "INV-001",
      "InvoiceDate": "2026-01-15",
      "Amount": "1250.00",
      "FileName": "invoice-001.pdf"
    },
    {
      "InvoiceNumber": "INV-002",
      "InvoiceDate": "2026-01-20",
      "Amount": "980.00",
      "FileName": "invoice-002.pdf"
    }
  ],
  "Errors": []
}

Configuration Fields

Field Name	Type	Required	Description
OneFileAtATime	Boolean	No	When `true`, each input file is sent in a separate AI request. Extracted rows receive a `FileName` column. When `false`, all files are sent together in one request. Default is `false`.
Column Map	Object Array	No	List of target columns and aliases. `Column` is the output column to extract. `Alias` describes alternate names that may appear in the document. Used to build the extraction objective and generated response schema.
Prompt	Text	Yes	User instructions for extraction. This is inserted into the activity’s AI table-extraction template.
StructuredOutput	Boolean	No	When `true`, the activity uses the provided `ResponseSchema`. When `false`, it builds a schema from `Column Map`.
ResponseSchema	JSON	Conditional	Custom response schema used when `StructuredOutput` is `true`. The schema should return data under a `Response` property if the default parser is expected to produce rows.

Conditional Field Rendering Rules

ResponseSchema is shown when StructuredOutput is true.

Sample Configuration

Scenario 1: Extract Invoices With Generated Schema

Field	Value
OneFileAtATime	`true`
Column Map	`InvoiceNumber = Invoice No, Bill No`; `InvoiceDate = Date, Bill Date`; `Amount = Total, Grand Total`
Prompt	`Extract invoice header fields from the attached files.`
StructuredOutput	`false`

Scenario 2: Use Custom Structured Output

Field	Value
OneFileAtATime	`false`
Prompt	`Extract all line items from the purchase order.`
StructuredOutput	`true`
ResponseSchema	`{ "type": "object", "properties": { "Response": { "type": "array", "items": { "type": "object" } } }, "required": ["Response"] }`

Sample Output

{
  "Data": [
    {
      "InvoiceNumber": "INV-001",
      "InvoiceDate": "2026-01-15",
      "Amount": "1250.00"
    }
  ],
  "Errors": [],
  "AdditionalResponse": "{\"Response\":[{\"InvoiceNumber\":\"INV-001\",\"InvoiceDate\":\"2026-01-15\",\"Amount\":\"1250.00\"}]}"
}