That reply means the similarity search found no chunk scoring above the chat's threshold — usually a tuning issue, not missing content. Check, in order:
- Is the content actually indexed? Open the embedding under Wyregraf → All Embeddings and confirm its chunks are indexed; if you recently edited the content, Reindex it.
- Namespace scope. If the chat is scoped to a namespace, the embedding must carry the same namespace, or the chat will never see it.
- Lower the Cosine Similarity Threshold on the chat's Restrictions tab. Content phrased differently from the question can score just under a strict threshold. Turn on Tuning in the embeddings profile to see each match's actual score while you calibrate, then set the threshold just below where good matches land.
- Raise Top K slightly, so more matches are considered per question.
- For questions phrased very differently from your prose, add Summaries & Questions embeddings — they're designed for exactly that gap.
Note that follow-up questions can also be answered from the conversation's own history, so a "no results" on the first question may resolve itself once the topic is established.
