Very large files

[ichoosetoaccept]

It appears as though Claude Coder (or is it kodu-coder?) is currently not smart enough to handle very large files.

I was working on a documentation project that involves trying to get a large swagger.json file (~24MB) into an intersphinx compatible format so I can use doc2dash on it to turn it into a Dash docset.

As long as Claude Coder worked with "normal sized" files (measured in KBs, not MBs) in the project it worked fine.

Whenever it tried to look at the large swagger.json or the generated index.html (also quite large), however, it would fail with "There was an issue with the format or content of your request."

Once it reaches this state it is unrecoverable and the only option left is to "Start New Task", try to summarize where we got stuck and continue starting fresh.

[matannahmani]

Thanks for the comment we are aware of that and it should be resolved in #11
I'm aware of issues with compressing tokens correctly, but in your specific use case, it is even more challenging as one message can fill the entire context window.
One approach we can take is splitting the file into chunks and adding a layer of RAG on top, this is probably one of the easier solutions to implement but it does require further thinking to come up with a more sustainable idea.
Do you mind sharing the letter count of the file you're uploading?