AI Extract

Choose the model that best fits your extraction needs. The available options are determined by the configured LLM Connector and may include models such as GPT-4o, GPT-4 Turbo, or GPT-3.5 Turbo.

• The selected model determines the language understanding, output quality, and cost of the extraction.
• Some features, such as structured output, may only be available on certain models.
• For best results, use the most capable model available for your use case.

Example:
Set to gpt-4o for high-accuracy extraction on complex documents.

Can be one of the following types:

Use this property to fine-tune how the LLM responds to extraction prompts. Parameters may include:

• Temperature: Controls randomness/creativity of the output (higher values = more creative).
• Max Tokens: Limits the length of the response.
• Top P, Frequency Penalty, Presence Penalty: Advanced controls for output diversity and repetition.

If left blank, default parameters from the LLM Connector will be used.

Tip:
Lower the temperature for more deterministic, repeatable results. Increase it for more flexible extraction in ambiguous cases.

Provides custom instructions to influence how the LLM interprets and extracts data from the document.

• Use instructions to clarify the extraction task, specify constraints, or provide examples.
• Well-crafted instructions can improve accuracy, especially for complex or ambiguous documents.

The Instructions property provides global guidance for the LLM, setting the overall context, rules, or high-level goals for the extraction operation. These instructions are included as a top-level message in the LLM prompt, shaping the model's general approach to the document.

In addition, the Description property on each individual Data Element (such as a Data Field, Data Table, or Data Section) is incorporated into the extraction schema and provided to the LLM as field-specific guidance. Each Description is embedded alongside the schema definition for its corresponding field, table, or section, allowing you to specify requirements, formats, or clarifications unique to that data element.

How They Work Together:

• Use the Instructions property to describe the overall extraction goal, document context, or global rules (e.g., "Extract all invoice dates and totals. Ignore line item details.").
• Use the Description property on each Data Element to provide detailed, field-level guidance (e.g., "The invoice date is typically found near the top right and should be in mm/dd/yyyy format.").
• Both properties are presented to the LLM: Instructions as a system/user message, and each Description as part of the schema for its field.
• This dual approach maximizes extraction accuracy: Instructions set the global context, while Description refines extraction for individual fields, tables, or sections.

Examples:

• Instructions: "Extract all invoice dates and totals. Ignore line items which represent returns."
• Field Description: "The invoice date is typically found near the top right and should be in mm/dd/yyyy format."
• Table Description: "List all line items, including description, quantity, and unit price."

Can be one of the following types:

The quoting method determines what portion of the document is provided to the LLM for context.

• Choose a quoting strategy that balances accuracy, cost, and prompt size.
• Options may include quoting the entire document, only relevant sections, or using custom logic.
• If left blank, the entire document content is included by default.

Tip:
For large documents, restrict quoting to only the necessary content to reduce token usage and improve focus.

Use this property to limit extraction to a subset of fields, tables, or sections.

• If left empty, all visible, non-computed descendant Data Elements are included.
• Add specific Data Fields, Data Tables, or Data Sections to focus extraction and reduce prompt complexity.
• Useful for incremental extraction or when only certain data is needed from a document.

Example:
Include only the "Invoice Date" and "Total Amount" fields for a targeted extraction.

• When enabled, instructs the LLM to return data in a strict JSON format matching the extraction schema.
• Supported by GPT-4o, GPT-4o mini, GPT-4 Turbo, and all GPT-3.5 Turbo models newer than gpt-3.5-turbo-1106.
• When disabled, uses OpenAI's legacy "JSON mode" or a best-effort parsing approach for non-OpenAI models.
• Structured output improves reliability and parsing accuracy, but may not be available on all models.

Tip:
Enable this option whenever possible for the most robust extraction results.

The 'Name' property identifies this fill method within the list of fill methods attached to a Data Model, Data Section, or Data Table.

• Use a clear, descriptive name to indicate the purpose or strategy of the fill method (e.g., "Run Child Extractors", "External Lookup", "Custom Calculation").
• The name is displayed in configuration UIs and logs, and is used to distinguish between multiple fill methods on the same container.
• The name is also used by the Fill Descendants command to identify which fill methods to execute during extraction.

The 'Trigger' property allows you to specify a condition under which this fill method will execute.

• If left empty, the fill method always runs.
• Enter a Boolean expression referencing data values, variables, or context to control execution.
• This enables conditional extraction logic, such as running a fill method only for certain document types or when specific data is present.

Example:
@DocumentType == "Invoice" && @TotalAmount > 0

If 'Overwrite' is set to true, this fill method will replace any existing values in the target data elements.

• Set to false to only populate fields and collections that are currently empty.
• Use this option to control whether extraction results should always be refreshed, or only filled in when missing.
• This is especially important when multiple fill methods are configured, or when data may be pre-populated from other sources.