AI assistant

The AI assistant helps with one PostgreSQL or TimescaleDB instance: it writes a query from a description, writes row-level security policies for your tables, explains why a statement failed, and explains an EXPLAIN plan. Open it from the instance’s Assistant tab or Database tools → AI assistant.

Turn it on

AI features are off until an owner or admin turns them on in Settings → AI (multi-factor authentication required). The same setting controls the SQL editor’s explain-and-fix. They are never available for instances on secure (HIPAA) placements, whatever the setting says.

What the model sees

Sent	Never sent
Your request as you typed it	Table rows or query results
Names of schemas, tables, columns, types, enums, indexes, foreign keys, policies and group roles	Row counts and table sizes
Column defaults, policy expressions and index definitions, with every literal value replaced by ?	Comments on tables and columns
The statement, error and plan you include, with literal values replaced by ?	Credentials, connection strings, login role names and grants
Advisor findings (lint names and the objects they name)	Anything from secure (HIPAA) placements

Databasezy keeps an audit entry per request with the task, the model and a fingerprint of your request, never its text. The provider is Anthropic (Claude).

Nothing runs by itself

• Queries are shown with a label: read-only, or changes data or schema. Open in SQL editor puts the statement in the editor; it runs only when you run it, and a statement that writes needs a read-write console session.
• Policies come back as schema changes. Review and apply shows the exact statements (enable row-level security, create or replace policies) and applies them only with your confirmation, like any change from the table editor or the policy editor.

Limits

Requests per organization per day depend on the plan (see Pricing and limits); short bursts are rate limited. When no AI provider is configured on a deployment the assistant shows as unavailable.