Ask HN: Is anyone successfully integrating LLMs with their entire codebase?

#1

Post by roddylindsay » Mon, Jul 15, 2024, 7:20 PM UTC

My team is using tools like Copilot, Phind and Coderabbit, but these seem mostly useful for code suggestions and feedback related to individual files.

Is anyone successfully using tools which, via large context windows/RAG/other IR techniques, are able to generate or review code with complex dependencies across many different files and code areas (i.e. db, backend, API, frontend)? Or is this still a ways out?

Re: Ask HN: Is anyone successfully integrating LLMs with their entire codebase?

#2

Post by wizzard0 » Mon, Jul 15, 2024, 7:38 PM UTC

Aider is one tool which works on medium sized repos for me. Make sure to use claude3.5-sonnet or gpt4o, other LLMs are not yet up to the task with the context size required

Re: Ask HN: Is anyone successfully integrating LLMs with their entire codebase?

#3

Post by Oras » Mon, Jul 15, 2024, 7:53 PM UTC

wizzard0 wrote: post #2

Aider is one tool which works on medium sized repos for me. Make sure to use claude3.5-sonnet or gpt4o, other LLMs are not yet up to the task with the context size required

I watched some videos about aider today, and some claimed it works well with DeepCoder V2. Haven't tried it yet, but interested to see if you had experience with it?

Re: Ask HN: Is anyone successfully integrating LLMs with their entire codebase?

#4

Post by wizzard0 » Mon, Jul 15, 2024, 8:37 PM UTC

Oras wrote: post #3

wizzard0 wrote: post #2

Aider is one tool which works on medium sized repos for me. Make sure to use claude3.5-sonnet or gpt4o, other LLMs are not yet up to the task with the context size required

I watched some videos about aider today, and some claimed it works well with DeepCoder V2. Haven't tried it yet, but interested to see if you had experience with it?

Nope, have yet to find a deepseek v2 provider which doesn’t log my prompts