Skip to content
Alation Agent Studio Documentation

Data Product Generator

The Data Product Generator is an interactive agent that turns relational database query usage, BI usage, and catalog metadata into well-scoped, fully enriched, chat-ready data product proposals — and, when you approve, creates them. It opens with a short clarifying exchange, runs an evidence pipeline, presents a ranked shortlist for your approval, then loops through each approved cluster to create and enrich a complete draft.

How it works

Section titled “How it works”
  1. Asks how you want to work: default pipeline across a datasource/schema, specific tables, a single product, or enrichment of an existing product
  2. Identifies candidate table clusters from relational database query usage and catalog metadata
  3. Reconfirms each cluster against BI usage data and per-table usage detail
  4. Synthesizes tables into coherent clusters — tables queried together or feeding the same BI reports become one cluster; disjoint-usage tables become separate ones
  5. Ranks clusters by impact and presents the shortlist for your approval
  6. Creates a draft data product for each approved cluster, then enriches it completely before moving on

The agent never creates anything until you have approved the shortlist.

Input parameters

Section titled “Input parameters”

Required:

  • message (string): Your natural-language request. Examples:
    • “Generate data products for the SALES schema”
    • “Enrich the existing Orders product”
    • “Create a data product from the upstream tables behind the Revenue dashboard”
    • “Generate eval cases for the Orders product from its query logs and upstream BI reports”

Optional:

  • datasource_id (integer): Scope the evidence pipeline to a single datasource instead of scanning all active sources

Output format

Section titled “Output format”

The agent produces thinking, tool call, tool return, and text blocks as it works. At each gate (shortlist approval, metric approval, eval-case approval, promotion to ready) it stops and waits for your response before proceeding. The final message is a per-product status summary: created / enriched / ready / failed.

Enrichment checklist

Section titled “Enrichment checklist”

Every created product is enriched to the following before the agent marks it chat-ready:

  • Columns — all columns with name, type, and description; never an empty schema
  • Documentation links — each record set links back to its catalog page
  • Delivery system — database connection URI copied exactly from the catalog configuration
  • Relationships — observed joins between member tables
  • Metrics — proposed KPIs derived from recurring query patterns, added only after your explicit approval
  • Descriptions — product and per-column descriptions grounded in real usage, not boilerplate
  • Example questions — 5–10 natural-language starter prompts derived from observed query intent and BI report measures
  • Eval cases — question/SQL pairs, created and run only after your approval; failures are used to fix the spec before promotion
  • Chat-ready status — set only after the spec is complete, evals have passed, and you have approved promotion

Behavior notes

Section titled “Behavior notes”
  • One pipeline whether you want one product or many — the cluster count is an output of the evidence pipeline, not an upfront assumption
  • Checks for existing products before creating; prefers updating a draft over creating a duplicate
  • Each product in a batch is built independently — a failure on one does not abort the rest
  • Gaps in BI usage data prompt a question rather than being treated as confirmation that no consumers exist
  • Individual query details and user identities from usage data are never exposed
Section titled “Related”