VectorAPI

VectorAPI is a service for managing vector collections and performing vector similarity queries using a PostgreSQL vector database with the pgvector extension. Utilizes fastapi for the HTTP API, pgvector and SQLAlchemy for the vector database side and relies on pytorch for computing embeddings.

Getting started

Existing database

To get started with the VectorAPI, run:

docker run -p 8889:8889 -e DB_URL=postgresql+asyncpg://<user>:<password>@<host>:<port>/<dbname> grafana/vectorapi

New database

You can bring up a postgres database (ankane/pgvector) and vectorapi instance using docker compose:

docker compose up --build

To populate the local DB instance with test data from HuggingFace (see Grafana public datasets) run:

make populate-db

Making requests

See API docs for more details.

Embedding text

curl -X POST "http://localhost:8889/v1/embeddings" \
    -H "Content-Type: application/json" \
    -d '{"input":"I enjoy taking long walks along the beach.", "model":"BAAI/bge-small-en-v1.5"}'

Adding a vector to a collection

1. Create a collection

curl -X POST "http://localhost:8889/v1/collections/create" \
    -H "Content-Type: application/json" \
    -d '{"collection_name":"my_collection", "dimension":384}'

2. Add a vector to the collection

curl -X POST "http://localhost:8889/v1/collections/my_collection/upsert" \
    -H "Content-Type: application/json" \
    -d '{"id":"abc1", "metadata":{"key":"value"}, "input":"I enjoy taking long walks along the beach."}'

Vector search

curl -X POST "http://localhost:8889/v1/collections/my_collection/search" \
    -H "Content-Type: application/json" \
    -d '{"input":"beach walks"}'