Extraction

Extraction uses a template to ask a model for structured values from project records or full text.

Extraction tab showing template selection, field cards, scope settings, text source, model preset, and Run extraction button.
The left side manages the template and fields. The right side chooses the run scope, text source, model preset, and run options.

Template controls

Template dropdown

Choose the extraction template. A template defines the fields the model will fill.

Add template (upload)

Imports a template file into the project. Use this when reusing a template from another project.

Edit template

Opens the template editor for the selected template. Use this before running when field instructions need correction.

New template

Creates a fresh template for a new extraction plan.

Hide

Temporarily hides a field from the next run without deleting it from the template.

Field details

Shows each field name, type, key, null handling, guidance, and example output.

Extraction field details showing field keys, types, guidance, and example JSON values.
Good extraction fields are specific, answerable, and explain what evidence should be used.

Run controls

Choose the record scope, text source, model preset, and whether to skip rows with existing values. The model preset can use the default extraction model, inherit an agent model, or use a saved extraction preset from Settings. The selected model is recorded with the extraction output.

Configure extraction-capable models in Settings: Runtime and model presets before running extraction across a project scope.

  • Scope chooses which rows to extract from.
  • Text source chooses what text the model sees, such as title and abstract or full text when available.
  • Model preset chooses the runtime/model used for extraction.
  • Skip rows with existing values prevents overwriting completed rows unless you intentionally rerun them.
  • Run extraction queues the extraction job.

Edit the template

The template editor lets you revise the system prompt, format instructions, and fields. Use concise field guidance and include examples when the expected answer shape could be ambiguous.

Extraction template editor modal with template name, system prompt, format instructions, field settings, and save buttons.
The template editor is where you define what the extraction model should return and how each field should be interpreted.

Review outputs

Extraction results become project columns that can be viewed in Screening, used by the agent, exported, and included in reports. Open Jobs to check progress or errors for long extraction runs.

For reliable extraction, make each field answerable from the supplied document and avoid asking one field to do several unrelated jobs.

Next: Screening