Databases

PostgreSQL requirements

The RAG store relies on PostgreSQL with these extensions:

• vector (pgvector)
• zhparser (full-text search config zhcn_search)
• hstore (token cache)

Schema creation

There are two paths:

1. Automatic (recommended): The ingestion pipeline (rag-cli or /api/rag/upsert) calls store.EnsureSchema, which creates the documents table, indexes, and the zhcn_search text search config.
2. Manual: sql/schema.sql creates a similar schema but uses pg_jieba and jieba_search. If you use this file, make sure your query path and text search configuration align with the runtime code.

Tables

• documents: embeddings, metadata, and tsvector for hybrid retrieval.
• cache_kv: unlogged cache table storing token validation results (hstore).
• admin_settings: stored via migrations in migrations/.

Connection configuration

Preferred config key:

• global.vectordb.pgurl: full connection DSN.

Alternative keys (used when pgurl is empty):

• pg_host, pg_port, pg_user, pg_password, pg_db_name, pg_sslmode