What is Schema Intelligence in Nova AI?

Schema Intelligence is Nova AI's automatic database learning system. It runs a 5-phase pipeline that reads your database catalog, profiles data values, discovers relationships between tables, generates AI-powered descriptions for every table and column, and stores the results as persistent knowledge. This pre-learned understanding allows Nova to generate accurate, validated SQL queries against your real schema.

Does Schema Intelligence work with all database types?

Schema Intelligence supports 9 SQL dialects: PostgreSQL, MySQL, SQL Server, MariaDB, SQLite, Oracle, Snowflake, BigQuery, and DuckDB. It also works with API connectors (Stripe, Shopify, HubSpot, Google Analytics 4, Google Search Console, ShipStation) by analyzing the DuckDB tables that store synced API data.

How long does the schema learning process take?

Schema catalog capture (Phase 1) typically completes in under 3 seconds for databases with up to 200 tables. The full 5-phase pipeline, including data sampling and LLM enrichment, takes 1-5 minutes depending on database size. Nova processes up to 10 tables concurrently to maximize speed.

Is my data safe during schema analysis?

Yes. The Query Streams Agent runs on your own network and handles all data sampling locally. Only aggregated schema metadata reaches the AIâ€”table names, column types, value distributions, and relationship mappings. Actual data rows never leave your infrastructure. The Agent uses outbound-only encrypted connections, requiring no inbound ports or firewall changes.

What are Nova's learning modes?

Nova offers four learning modes configurable per database: Auto (runs the full pipeline automatically on connection), Assisted (asks for confirmation before analysis), Manual (triggered per table via analysis buttons), and Off (disables Schema Intelligence entirely). Most users start with Auto mode.

Can Schema Intelligence detect relationships that aren't declared as foreign keys?

Yes. Phase 3 of the pipeline specifically targets implicit foreign key discovery. It analyzes column names, data types, and value overlaps to identify undeclared relationships between tables. This is critical because many production databases have relationships that exist in practice but aren't formally declared as FK constraints.

How does Schema Intelligence improve SQL accuracy?

Schema Intelligence prevents hallucinated column names by verifying every reference against your real schema. It ensures correct filter values through data profiling, maps accurate JOIN paths via relationship discovery, generates dialect-specific syntax for your database engine, and validates every query through SQLGlot syntax checking before execution.

Does Schema Intelligence work with API data sources?

Yes. When you connect an API connector like Stripe, Shopify, or HubSpot through Query Streams, the data is synced into local DuckDB tables. Schema Intelligence analyzes these DuckDB tables the same way it analyzes any other database, learning the table structure, column types, and value patterns so Nova can generate accurate SQL against your API data.