The simpler the task I'm working on, the higher the likelihood it gets it right. This is where I get the most benefit from Copilot, as I already understand what I'm coding, why I'm coding it, and how it should appear. Copilot sometimes saves me the 5-30 seconds it would take to write it manually. Over the course of a day, these saved moments add up.
But I always found myself turning to ChatGPT for help writing SQL. I would 1) run a query to get the schema. 2) Feed schema to ChatGPT. 3) Ask GPT to write my query. 4) Paste query into client. Solid results, but the workflow was not great with lots of copy and paste and multiple apps.
I work at Hoopdev, an infrastructure access gateway and we happen to have a build-in web client. Therefore our language server has a REST API that we could hook to GPT4. But our first attempt failed. We tried the Ask AI box UX, which only solved half the problem: providing the schema as context. Not getting the way was still missing. The bad suggestions required another prompt to get fixed.
Therefore we went on to build the copilot experience [0], but the opposite of Github copilot happened: most suggestions are good. This is because the only context for writing SQL is the database schema and what you want to do, all the model has to do is translate your english query to another language (SQL) using a small context (schema), as opposed to full codebases.
Hoopdev's client is web-based, because of that you can try it in 15 seconds. After signing up you get a demo Postgres ready to be queried in the web [1]. If you don't wanna sign up, run it yourself and use your own OpenAI keys [2]. Please let me know what you think. any feedback is welcome. Is this helpful? Do you know/use a better alternative for SQL?
0. https://hoop.dev/blog/introducing-an-ai-powered-query-builde... 1. https://hoop.dev/start 2. https://hoop.dev/docs/deploy/AWS
