Ask in your own words and AisDrive compares your question for similarity with the fragments of all your indexed documents: it returns the piece of text that answers, the file it comes from, and its link to open it.
Searching by file name only works if you remember what you called it. AisDrive searches by what it says inside: by meaning, not by letter match.
When you send a file to be indexed in AisDrive, its text is extracted, split into fragments of about 400 words with 50 overlapping, and each fragment is converted into a 768-dimensional vector using the gemini-embedding-001 model. That vector is, in practice, the footprint of the fragment's meaning.
When you ask a question, your query is converted into a vector in the same space and compared with the fragments in your drive using cosine similarity. AisDrive retains the 6 closest fragments above a relevance threshold of 0.3, and also prioritizes those that literally contain the keywords from your query: this way, the search works both when you describe the idea with different words and when you remember the exact term.
Each retrieved fragment comes with its origin: file name, folder, relevance percentage, and links to view or download it. It does not return a list of file names that you would have to open one by one, but the specific paragraph and where it came from.
In AisDrive, this search is used from the chat: you write the question in natural language and the answer comes with cited sources, each navigable to its folder. The same retrieval is used by the AI assistant as a tool when it needs to consult your documents. There is (yet) no independent semantic search box in the file view.
Search by meaning
you find the document even if you don't use a single word from it, because the comparison is between sense vectors, not between text strings.
It returns the fragment, not the file
you see the paragraph that answers your question, with its relevance percentage, instead of a list of names that you would have to open manually.
Complete traceability
each result comes with the file name, its folder, and links to open or download it, so you can always check the source.
Hybrid when needed
if your query contains a literal term (a code, a proper name, a reference), the snippets that contain it rise in position even if their semantic similarity is low.
With noise filtering
snippets below 0.3 similarity are discarded, so the result is "no relevant content" instead of some random invented paragraph as a response.
Search across your entire drive at once
the query traverses snippets from all your indexed files in AisDrive, not just the folder you are in.
Index what you already have
Word, Excel, PowerPoint, txt, md, csv, json, html, and more are extracted locally at no reading token cost, and PDFs and images go through AI vision.
Private by default
the search is limited to your own content within your account; it does not mix files from other users.

Create your drive, index a few documents, and ask about an idea you partially remember. AisDrive will bring you the relevant fragment and the source file.