Advanced AI settings

When configuring an AI feature, expand the Advanced Settings section to adjust AI behavior. These settings are available in the AI Data Extraction task and other AI-powered features.

Parameters

These parameters control how the AI model generates and returns responses.

Temperature (0–2)

Temperature controls response randomness. Lower values produce more focused and deterministic output, while higher values introduce more variation.

  • 0–0.3 — Use for structured data extraction where consistency matters — for example, when extracting invoice numbers or dates.
  • 0.5–0.7 — General-purpose setting for most extraction tasks. Good for extracting summaries or descriptions.
  • 1.0+ — Produces more creative and varied output.

Top P (0–1)

Top P controls token selection diversity by limiting the pool of tokens the model considers, and works alongside temperature.

  • 0.1 — Only the most probable tokens are considered. Use for highly predictable output like extracting IDs or amounts.
  • 0.5–0.9 — Allows moderate diversity. Suitable for general extraction tasks.
  • 1.0 — All tokens are considered. Use when the AI needs maximum flexibility, such as generating summaries.

Max tokens

Max tokens sets the maximum length of the AI response. Higher token limits increase the cost per request, as most AI providers charge based on the number of tokens consumed.

Stop sequences

Stop sequences are text patterns that stop AI response generation, preventing the model from producing unwanted content after the required data.

  • Setting --- as a stop sequence prevents the AI from continuing past a logical boundary in its response.
  • If your prompt asks the AI to return JSON, use } followed by a newline to prevent the model from appending extra text after the closing brace.
  • If your extraction prompt returns a numbered list, a double newline (\n\n) can prevent the AI from adding unsolicited commentary after the extracted fields.

Return to request detail

When enabled, the AI Data Extraction task processes in the background while users continue working. The extracted data becomes available once processing completes.

  • Enabled — Recommended for large documents or complex prompts that take longer to process.
  • Disabled — The user waits for the AI to finish before proceeding. Use for quick extractions where immediate results are expected.