PostgreSQL databases

When connecting to a PostgreSQL database you can use them for several different GenAI services, such as embeddings, document stores, chat history or analytics.

An alternative to running your own PostgreSQL database is to use the AlloyDB managed service on Google Cloud Platform.

Embedding

When connecting to a PostgreSQL database for vector embeddings, the pgvector extension needs to be installed within the database - refer to the documentation for the specific provider.

When you have the requisite details, then you need a username, password, ip of the PostgreSQL database and the database used to be put into a connection string and set to the PGVECTOR_CONNECTION_STRING environment variable - an example is shown below:

PGVECTOR_CONNECTION_STRING=postgresql://user:password@1.2.3.4:5432/database

To use within sunholo, you can use the pick_retriever() function to pull in the configuration dynamically according to the vector_name argument.

A configuration can be set that will send embeddings after chunking to the database.

An example config is shown below:

kind: vacConfig
apiVersion: v1
vac:
  sample_vector:
    llm: openai
    agent: langserve
    memory:
      - azure_postgres:
          vectorstore: "postgres"

This will then return the retriever via

retriever = pick_retriever("sample_vector")