Import unstructured data (text and images) into structured tables

Install this plugin in the same environment as Datasette.

datasette install datasette-extract

This plugin requires an OpenAI API key.

You can set this using the DATASETTE_SECRETS_OPENAI_API_KEY environment variable, or you can configure the datasette-secrets plugin to allow users to enter their own plugin and save it, encrypted, in their database.

This plugin provides the following features:

• In the database action cog menu for a database select "Create table with extracted data" to create a new table with data extracted from text or an image
• In the table action cog menu select "Extract data into this table" to extract data into an existing table

When creating a table you can specify the column names, types and provide an optional hint (like "YYYY-MM-DD" for dates) to influence how the data should be extracted.

When populating an existing table you can provide hints and select which columns should be populated.

Text input can be pasted directly into the textarea.

Drag and drop a PDF or text file onto the textarea to populate it with the contents of that file. PDF files will have their text extracted, but only if the file contains text as opposed to scanned images.

Drag and drop a single image onto the textarea - or select it with the image file input box - to process an image.

Users must have the datasette-extract permission to use this tool.

In order to create tables they also need the create-table permission.

To insert rows into an existing table they need insert-row.

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

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

Now install the dependencies and test dependencies:

pip install -e '.[test]'

To run the tests:

pytest