SearchThing Field Notes
Private search starts with a smaller trust boundary
SearchThing keeps parsing, indexing, and ranking in the browser, so a useful document search does not need a cloud copy of the documents it knows.
A private search tool starts with a question about where the index lives. If the documents stay in the browser, the search can stay close too.
Search the words and the meaning
BM25 is quick when the words line up. Local vectors help when the idea is right but the phrase is different. SearchThing hybrid mode combines both signals with reciprocal rank fusion so exact and semantic matches can share one result list.
Keep the index where the files are
Documents live in browser storage and embeddings use OPFS-backed persistence. The browser becomes the workspace instead of a pass-through to a hosted index. That makes the trust boundary easier to see and easier to reason about.
Start with a text search
Text results can arrive immediately. When the local embedding model is ready, semantic matches can sharpen the list without changing the trust boundary.
Good local search is not about pretending every result is certain. It is about showing enough of the path—from query to ranking signal—that you can decide whether the result is useful.