PostgreSQL × AI · Machine-readable entry point

AI documentation is not a long manual pasted into a prompt.

Use small, stable knowledge units. Declare the PostgreSQL version, schema, role, and execution boundaries, and make every answer traceable to verifiable evidence.

Open AI documentation llms.txt

context-contract.json

{  "database": "postgresql",  "major": 18,  "role": "app_readonly",  "statement_timeout_ms": 3000,  "required_evidence": ["schema", "SQLSTATE", "canonical_url"]}

Core boundaries for AI on PostgreSQL

Context contract

Declare version, role, schema, timeouts, and output.

Execution boundaries

Parameters, read-only mode, row limits, transactions, and audit.

Agent evaluations

Test correct outcomes and forbidden privilege violations.

Machine-readable formats

llms.txt

Site inventory and recommended retrieval entry points.

llms-full.txt

Full corpus entry point for offline ingestion.

Single-page Markdown

Single-page machine documentation with canonical URLs.

Humans remain part of the entry point

Understand, then automate

Build a database mental model first, then let agents work under explicit permissions, timeouts, and audit rules.

Start from zero