Most note-taking apps make you remember exactly what you wrote and how you titled it. Daneshbase doesn't.
Every doc in your wiki is indexed two ways: a full-text engine for exact-word matches, and a vector embedding for the meaning of what you wrote. When you search, both run in parallel and the results are fused — so a query for "how do I redact a PDF" finds the doc you titled "Hiding sensitive info in scanned receipts" even if you never wrote "redact" or "PDF" anywhere in it.
Ask in one language, find docs in another
The embedding model is multilingual. Search "Wochenplanung" — German for "weekly planning" — and you'll get back the English doc where you wrote about your Sunday review ritual. No translation step, no aliases to maintain.
Reranked when it matters
For lookups where picking the right doc matters more than getting a result fast, the search runs a reranker pass over the candidates. The reranker scores each result against the query directly instead of trusting the embedding's nearest-neighbour vote. Useful when an AI agent has to act on whatever doc you hand it.
No setup, no rules
Search just works on every doc, every revision, every locale. You don't author keywords, configure synonyms, or tag for retrieval — the indexer reads the markdown and the AI keeps it useful as your wiki grows.