This feature is available as a private beta to select customers. Please contact us if you are interested. Any feature described below may change or be removed at any time.
PopSQL harnesses the power of generative AI to provide a number of features to help provide you assistance across a number of tasks. While we aim to have the AI provide you meaningful information, any generated response may not be accurate or complete, and we recommend carefully reviewing any code it generates.
To power each of the below features, we use a third party API provided by OpenAI. Please make sure to review the disclaimer at the bottom of this page for additional information on what information we send to OpenAI's API.
On the left sidebar, you can open a chat with our AI assistant by clicking the AI icon. With the window open, you can ask it questions about how best to write a query, your schema, or whatever other SQL related questions you might have. When asking a question, some number of your most recently asked questions and their answers will be provided to the AI assistant for context in answering your current question.
Check out this quick demo showing some examples of using the AI Chat in PopSQL:
For the chat, there are three models that are available to be used:
gpt-3.5-turbo-32k- Default model, is slower than
gpt-3.5-turbo, but includes more previous questions and answers
gpt-3.5-turbo- The fastest model
gpt-4- Slower than
gpt-3.5-turbo, but can provide higher quality answers
- Each model has a maximum amount of context we can send it, so not all previously asked questions/answers may be sent to the model
- The assistant can only answer questions about schema with information shown in the tree, such as table and column names, primary keys, indexes, triggers, etc. Anything not shown in the tree (e.g. foreign keys, partitioning information, etc.) is not available to the assistant.
When running a query, if it fails to run due to a syntax error, we provide a "Fix errors" button that will attempt to automatically figure out how to fix your query, showing the changes as a diff. You can then choose to accept or reject the generated diff.
You can have our AI assistant generate a title or description for your queries. To do this, click the three dot menu and select the appropriate option.
To power each of the below features, we use a third party API provided by OpenAI. To help power these services, we may send the following information to the API:
- The title, description, and body of the query you are looking at
- The client type (e.g. postgresql, snowflake) and schema for the connection for the query you are looking at
- A small sample of the results for the last run (this can be disabled if you do not wish to have this sent)
Please review their API data privacy doc to understand what they can do with the information sent to their API.
For our own purposes for fine tuning these features, we may record the following data:
- Prompt sent to the API and its generated response
- Model used
- Metadata for the feature used (e.g. ID of query currently looking at, ID for chat)
Updated 3 months ago