Replies: 1 comment
-
|
I’d keep the source material in the bundle if it’s something the model is expected to cite or extract from directly. For PDFs, images, or video, I’d treat them as first-class references and store them under a stable That said, I’d separate the canonical source from derived text. So the bundle can point to the original asset, while also carrying extracted text or summaries where that helps retrieval. In practice, the useful rule is: if the document is part of the knowledge base, keep a stable pointer to it in the bundle; if it’s just supporting material, keep it external and reference it. That gives you the best of both worlds without bloating the bundle. |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
-
Wondering how the spec sees including references in the knowledge bundle. Some of my reference docs are obviously images or PDFs, or even .mov files that the llm has tools to extract content from. I think keeping the source content is valuable but I have been keepign it outside of the bunle . The spec seems to suggest keeping these potentially in the knowledge itself.
"Citation links MAY be absolute URLs, bundle-relative paths, or paths into a references/ subdirectory that mirrors external material as first-class OKF concepts."
Is this how others are handling this?
Beta Was this translation helpful? Give feedback.
All reactions