Configure AI on a table field

Configure the input and model

Prompt interpolation accepts a field title or internal name. Source-field actions track the selected source directly; prompt-driven auto-update depends on placeholders that the dependency planner can resolve.

• Choose a source field for actions that require one
• For prompt-based actions, reference record values with {{field title-or-name}} placeholders
• Optionally choose a per-field runtime model override from models compatible with the action
• For text actions that read attachments, choose a vision-capable model
• Save the field configuration before running generation

Manual and automatic regeneration

New AI configurations default to automatic regeneration off. Run an individual cell from its regenerate control, regenerate the configured AI fields for a record from the record menu, or choose Generate from an AI-enabled column menu.

When automatic regeneration is enabled, creating a record or changing a tracked dependency can schedule downstream AI fields in dependency order. Cyclic or missing dependencies cannot be treated as a reliable pipeline.

Verified batch boundary

A column generation request is bounded to 100 unique positive record IDs. Prompt-placeholder batches are materialized per record with client concurrency of three; other actions use the runtime batch endpoint after preparing the selected records.

The UI reports partial failures and keeps per-cell generation state. Review failed rows and rerun only the records that need it.

Cost and model guidance

Model availability and pricing come from the connected provider and can change. Select a compatible model from the current runtime catalog and review usage with the provider or the applicable Kition plan surface.

Kition 0.1.41 does not support the old documentation’s fixed per-row prices, hard-coded model recommendations, Settings → Usage per-field ledger, provider-routing promise, prompt-cache invalidation menu, or “show last prompt” cell action.