Description
AI Content Enrichment is a family of opt-in behaviors for summaries, title proposals, entity extraction, quality scoring, question generation, and action-item extraction.
Each behavior receives the row through the shared automation context, prepares bounded plaintext, checks its own configuration and provider availability, and validates structured model output before proposing a mutation. The behaviors remain separate so a Dotabase can enable only the enrichment it needs and route every result to a schema-declared property. Their effects then use the normal behavior apply path rather than bypassing row validation.
Selling Points
- Converts free-form content into useful structured signals.
- Writes only schema-declared target properties.
- Uses versioned prompts, cost caps, and strict output contracts.
- Lets teams enable summaries, entities, questions, or actions independently.
- Turns enrichment failures into explicit skips or deferrals instead of corrupting rows.
User Story
As a knowledge worker, I turn a long note into a summary, entities, questions, and follow-up work without copying content between tools. I enable the relevant bindings on the Dotabase and choose target properties that fit its schema. When a row changes, the selected behaviors analyze the same source content and return typed proposals. I walk away with reviewable structured signals beside the original note, while the source remains intact and ordinary property rules still govern every write.
Extension Surface
Configurable: each enrichment has its own binding, thresholds, and target-property settings.
packages/dotabases/src/behaviors/auto-summary/index.ts demonstrates the extension boundary: configuration selects the target and execution policy, but the behavior definition and output contract stay platform-owned. Additional shipped enrichments follow the same binding shape; customers configure them rather than loading arbitrary prompt code.
Capabilities & Limits
- Capability: Executes the scoped workflow described above through the cited live implementation paths.
- Capability: Preserves typed validation and applicable tenant, schema, lifecycle, or service boundaries.
- Limit: Provider absence produces deferral rather than synthetic results. Title suggestions require explicit acceptance unless a declared suggestion field is used. The family can summarize content, extract entities and action items, and generate other bounded signals without granting the model authority over undeclared fields. Cost gates and schema checks protect the durable row. Results remain probabilistic, depend on configured prompt and model providers, and should not be treated as verified facts. Enrichment also does not replace human judgment for titles or consequential workflow decisions.
Implementation Map
packages/dotabases/src/behaviors/auto-summary/index.tspackages/dotabases/src/behaviors/entity-extraction/index.tspackages/dotabases/src/behaviors/action-item-extraction/index.ts