Python

SereneDB works with psycopg, Python's most popular PostgreSQL adapter. Both psycopg 3 and psycopg2 are supported.

Install

sh

pip install "psycopg[binary]"

Connect

python

import psycopg

conn = psycopg.connect("host=localhost port=7890 dbname=postgres user=postgres")

Create a table and insert data

python

with conn.cursor() as cur:
    cur.execute("""
        CREATE TABLE IF NOT EXISTS articles (
            id INTEGER PRIMARY KEY,
            title TEXT,
            views INTEGER
        )
    """)
    cur.execute(
        "INSERT INTO articles VALUES (%s, %s, %s)",
        (1, "Introduction to Vector Search", 4200)
    )
    conn.commit()

Query

python

with conn.cursor() as cur:
    cur.execute("SELECT title, views FROM articles ORDER BY views DESC")
    for row in cur:
        print(row)

Using psycopg2

psycopg2 is also supported:

sh

pip install psycopg2-binary

python

import psycopg2

conn = psycopg2.connect("host=localhost port=7890 dbname=postgres user=postgres")

with conn.cursor() as cur:
    cur.execute(
        "INSERT INTO articles VALUES (%s, %s, %s)",
        (1, "Introduction to Vector Search", 4200)
    )
    conn.commit()

Cleanup

python

conn.close()

Differences from PostgreSQL

• A conn.transaction() block inside another one fails, because the inner block needs a savepoint and SereneDB has no SAVEPOINT. Keep one transaction per unit of work.
• Named cursors (conn.cursor(name=...)) fail, because DECLARE ... CURSOR is not supported. Read large results with cur.stream(query) in psycopg 3, or page through them with LIMIT and a key.
• LISTEN and NOTIFY are not supported.
• psycopg sends a Python list of floats as DOUBLE[]. Cast it to the column type when you compare it with a vector column: ORDER BY embedding <-> %s::FLOAT[768].

LangChain

For LangChain applications, SereneDB ships a vector store integration that builds on psycopg 3: see langchain-serenedb.