Hacker News Reader

Comment by ukuina | Hacker News Reader

It's fascinating to see how different frameworks are dealing with the problem of populating context correctly. Aider, for example, asks users to manually add files to context. Claude Dev attempts to grep files based on LLM intent. And Continue.dev uses vector embeddings to find relevant chunks and files.

I wonder if an increase in usable (not advertised) context tokens may obviate many of these approaches.

View on HN

I've been frustrated with embedding search approaches, because when they fail, they fail opaquely -- I don't know how to iterate on my query in order to get close to what I expected. In contrast, since repogather merely wraps your query in a simple prompt, it's easier to intuit what went wrong, if the results weren't as you expected.

> I wonder if an increase in usable (not advertised) context tokens may obviate many of these approaches.

I've been extremely interested in this question! Will be interesting to see how things develop, but I suspect that relevance filtering is not as difficult as coding, so small, cheap LLMs will make the former a solved, inexpensive problem, while we will continue to build larger and more expensive LLMs to solve the latter.

That said, you can buy a lot of tokens for $150k, so this could be short sighted.

I am really happy with how Aider does it as it feels like a happy medium. The state of LLMs these days means you really have to break the problem down to digestible parts and if you are doing that, its not much more work to specify the files that need to be edited in your request. Aider can also prompt you to add a file if it thinks that file needs to be edited.

I haven't looked into this but do any of them use modern IDE code inspection tools? I'd think you would dump as much "find references" and "show definition" outputs for relevant variables into context as possible.

Aider also uses AST (tree sitter), it creates repo map using and sends it to LLM.

I like this approach a lot, especially because it's not opaque like embeddings. Maybe I can add an option to use this approach instead with repogather, if you are cost sensitive.

How does Aider (cli) compare to Claude Dev (VSCode plugin)? Anyone have a subjective analysis?

And how does repogather do it? From the README, it looks to me like it might provide the content of each file to the LLM to gauge its relevance. But this would seem prohibitively expensive on anything that isn't a very small codebase (the project I'm working on has on the order of 400k SLOC), even with gpt-4o-mini, wouldn't it?

repogather indeed as a last step stuffs everything not otherwise excluded through cheap heuristics into gpt-4o-mini to gauge relevance, so it will get expensive for large projects. On my small 8 dev startup's repo, this operation costs 2-4 cents. I was considering adding an `--intelligence` option, where you could trade off different methods between cost, speed, and accuracy. But, I've been very unsatisfied with both embedding search methods, and agentic file search methods. They seem to regularly fail in very unpredictable ways. In contrast, this method works quite well for the projects I tend to work on.

I think in the future as the cost of gpt-4o-mini level intelligence decreases, it will become increasingly worth it, even for larger repositories, to simply attend to every token for certain coding subtasks. I'm assuming here that relevance filtering is a much easier task than coding itself, otherwise you could just copy/paste everything into the final coding model's context. What I think would make much more sense for this project is to optimize the cost / performance of a small LLM fine-tuned for this source relevance task. I suspect I could do much better than gpt-4o-mini, but it would be difficult to deploy this for free.

Continue.dev approach sounds like it would provide the most relevant code?

Embeddings are actually generally not that effective for code.

This is what I've found. It's so hard to do embeddings correctly, while it's so easy to search over a large corpus with a cheap LLM! Embeddings are also really inscrutable when they fail, whereas I find myself easily iterating if repogather fails to return the right group of files.

Of course, 'cheap' is relative -- on a large repository, embeddings are 99%+ cheaper than even gpt-4o-mini.

Reply on news.ycombinator.com