Automatic Contexting?

[cchance27]

I noticed something interesting when playing with some other tools, one of which was Trae, ya i know they're basically harvesting data from projects, for their model, but since i'm working on opensource i don't really care as much.

That said i noticed the first time i ran it i refrenced a file in context, and it took a good 15-20seconds before it started running for first question but it seems like when it finally submitted it had expanded the context automatically before submitting the first request....

Not to a bunch of files but to specific lines of files, and this got me thinking, i don't know how their building that context (maybe via RAG or semantic search based on the main file i was adding to context?), but, wouldn't it be nice if we could generate a call graph for a file before we submit a query and then use that call graph to grab the specific functions and lines to include in context instead of dumping the entire codebase to context.

Theoretically it could be possible to limit the depth of the call graph to include during traversal...

What do you guys think?

[NaccOll]

codebase in progress