Defog privacy friendly architecture means that we never get access to your data. Defog only has access to your metadata (tables names, column names, column types, and column descriptions) – and uses these to convert natural language questions into machine-executable queries. These queries are then executed on your infrastructure, ensuring that your data never leaves your servers.

This makes Defog ideal for industries where privacy is of the essence – healthcare, fintech, insurance. But we take privacy seriously beyond just our architecture. We have taken measures to ensure that your metadata is always secure, the code generated is never harmful, and you can add filters to address take access issues.

Protecting your metadata

We make sure that your metadata cannot be reached by unauthenticated users. We have taken a number of steps to ensure this:

Preventing generation of harmful queries

We have hard checks in place to ensure that the queries generated are queries that only read the data – and never update, delete, or otherwise modify the data in your database. Our model will never create a query that modifies the data in your database in any way.

Addressing data access issues

Our hard_filters feature makes sure that users cannot access rows or columns in your database that you don’t want them to. Case in point here!

1from defog import Defog
2
3defog = Defog()
4user_question = "can you show me the bank balance of all users?"
5answer = defog.run_query(
6    user_question,
7    hard_filters="show data only for the user_id 123"
8)
9# only data for the user_id `123` will be returned
10# regardless of what the question is