Skip to main content

Store and query embedding vectors in Datasette tables

Project description

datasette-embeddings

PyPI Changelog Tests License

Store and query embedding vectors in Datasette tables

Installation

Install this plugin in the same environment as Datasette.

datasette install datasette-embeddings

Usage

Adds an enrichment for calculating and storing OpenAI embedding vectors for content in a table.

Users get to select the embedding model and the template (e.g. {{ title }} {{ body }}) for the columns they would like to embed.

Embeddings are stored as binary values in columns in a new table called _embeddings_NAME, where NAME is the name of the original source table.

The vectors are stored in columns that match the name of the embedding model, for example emb_text_embedding_3_large_256 for the text-embedding-3-large-256 model.

If you do not configure an OpenAI API key users will be asked for one any time they run the enrichment.

You can set an API key with plugin configuration like this:

plugins:
  datasette-embeddings:
    api_key:
      $env: OPENAI_API_KEY

Then set the OPENAI_API_KEY environment variable before you start Datasette.

This plugin adds a "Semantic search against this table" table action item for tables with embeddings, but only if the API key environment variable has been configured as the key is needed to calculate embeddings for the user's search query.

Development

To set up this plugin locally, first checkout the code. Then create a new virtual environment:

cd datasette-embeddings
python3 -m venv venv
source venv/bin/activate

Now install the dependencies and test dependencies:

pip install -e '.[test]'

To run the tests:

pytest

The tests use captured examples of embedding APIs. The easiest way to re-generate these is to do the following:

  • rm -rf tests/cassettes to remove the previous recordings
  • export OPENAPI_API_KEY='...' to set an OpenAI API key
  • pytest --record-mode once to recreate the cassettes

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

datasette-embeddings-0.1a3.tar.gz (12.5 kB view hashes)

Uploaded Source

Built Distribution

datasette_embeddings-0.1a3-py3-none-any.whl (11.3 kB view hashes)

Uploaded Python 3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page