Chat Agent schema design recommendations Innovation Release

Hybrid Manager (HM) Chat Agent can analyze your table schemas and query workload and suggest improvements to your schema design. Recommendations are evidence-based and grounded in Postgres and EDB best practices.

Recommendations are delivered conversationally — ask Chat Agent about a specific table and it returns actionable suggestions. Schema design recommendations don't appear in the Recommendations tab; they're available only through Chat Agent.

Prerequisites

• The target cluster is registered with HM. Both HM-managed and self-managed clusters are supported.
• Monitoring is enabled on the cluster. Chat Agent reads query statistics and schema metadata through HM's monitoring data, which is available for both managed and self-managed clusters. Without monitoring data, recommendations fall back to schema-only analysis.

Recommendation types

Chat Agent focuses on DDL-level improvements for both managed and self-managed clusters — changes to how your data is defined and stored, rather than to the queries that access it. It applies a Postgres and EDB best-practices catalog, so the suggestions you get back depend on the table and its workload. Common areas include:

• Normalization and structure — flags issues such as repeating groups, partial dependencies, or over-denormalization that may be affecting performance or maintainability.
• Data types — suggests more appropriate types for a column, for example TIMESTAMPTZ over TIMESTAMP, NUMERIC for money, or TEXT over CHAR(n)/VARCHAR(n).
• Constraints and keys — recommends primary keys, foreign keys, NOT NULL, CHECK, and UNIQUE constraints where they're missing or could be tightened.
• Indexing — identifies missing indexes for your access paths (such as unindexed foreign-key columns) and the appropriate index type, and flags unused or redundant ones.
• Partitioning — identifies tables that would benefit from range, list, or hash partitioning based on query patterns and data distribution.
• Workload-specific tuning — adjustments for update-heavy, insert-heavy, or upsert-oriented tables, such as fillfactor settings or separating hot and cold columns.
• EDB extensions — where it adds value, recommends EDB extensions for the use case, for example the EDB Advanced Storage Pack table access methods for analytical or append-only workloads.

For curated, auto-applicable index recommendations, see Recommendations and advisories.

How schema recommendations work

When you ask for schema recommendations, Chat Agent:

• Reads the table's DDL (column definitions, constraints, indexes, and storage settings).
• Retrieves the top queries running against the table. By default it looks at the last 6 hours; you can specify a different window in your prompt.
• Applies a Postgres and EDB best-practices catalog to the schema and query data.
• Returns specific, evidence-based recommendations with an explanation for each.

If there's no query workload in the specified window, Chat Agent falls back to schema-only recommendations derived from the table definition alone.

Example prompts

• Suggest schema improvements for table public.orders on customers-prod, database app.
• Review the schema for public.line_items on orders-prod using the last 24 hours of query data.
• Are there any partitioning opportunities on public.events in analytics-db?

Limitations

• Recommendations cover DDL-level improvements only. Query-level (DML) improvements are out of scope.
• Schema recommendations aren't auto-applied. Unlike index recommendations, you must implement them manually or through your own deployment pipeline.
• Recommendations aren't displayed in the Recommendations tab or included in the cluster health score.
• Accuracy improves with more query data. For newly created tables or clusters with little workload history, recommendations are based on schema structure alone.