rembrembdocs
← supabaseoriginal source ↗

Python client

Manage unstructured vector stores in Postgres.

Supabase provides a Python client called vecs for managing unstructured vector stores. This client provides a set of useful tools for creating and querying collections in Postgres using the pgvector extension.

Quick start#

Let's see how Vecs works using a local database. Make sure you have the Supabase CLI installed on your machine.

Initialize your project#

Start a local Postgres instance in any folder using the init and start commands. Make sure you have Docker running!

1# Initialize your project2supabase init34# Start Postgres5supabase start

Create a collection#

Inside a Python shell, run the following commands to create a new collection called "docs", with 3 dimensions.

1import vecs23# create vector store client4vx = vecs.create_client("postgresql://postgres:postgres@localhost:54322/postgres")56# create a collection of vectors with 3 dimensions7docs = vx.get_or_create_collection(name="docs", dimension=3)

Add embeddings#

Now we can insert some embeddings into our "docs" collection using the upsert() command:

1import vecs23# create vector store client4docs = vecs.get_or_create_collection(name="docs", dimension=3)56# a collection of vectors with 3 dimensions7vectors=[8  ("vec0", [0.1, 0.2, 0.3], {"year": 1973}),9  ("vec1", [0.7, 0.8, 0.9], {"year": 2012})10]1112# insert our vectors13docs.upsert(vectors=vectors)

Query the collection#

You can now query the collection to retrieve a relevant match:

1import vecs23docs = vecs.get_or_create_collection(name="docs", dimension=3)45# query the collection filtering metadata for "year" = 20126docs.query(7    data=[0.4,0.5,0.6],      # required8    limit=1,                         # number of records to return9    filters={"year": {"$eq": 2012}}, # metadata filters10)

Deep dive#

For a more in-depth guide on vecs collections, see API.

Resources#