Show HN: LLM output to editor edits, A step by step guide

4 points

3 years ago

Hi there! My co-founder, Sandeep, and I at CodeStory have been busy hacking an in-line editor agent and want to share our learnings and some of the technical challenges along the way.

We use a mix of tree-sitter queries and LLM prompting to get the in-line code generation to work.

We wanted to get this to work with GPT3.5 because of 1. latency, 2. OSS models which are more capable than GPT3.5 can be dropped in as a replacement.

I hope you find the tech deep dive useful! Do let me know if you have more questions about this, happy to answer any and all of them.

If you like what we are doing, please try out our editor here: https://github.com/codestoryai/binaries/releases/tag/1.84.0....

and you can join our discord to follow along on our journey: https://discord.gg/FdKXRDGVuz

Thanks!