From upload to answer.
Five steps. The slow part, reading, is left to your LLM, and it reads very little.
Five steps. No vectors.
- 01Any text
Upload
Docs, PDFs, Office files, e-books, zips, audio or a GitHub repo.
- 020.1 s
Index
A trigram index of every line, and a brief of every file.
- 03Plain words
Ask
A question in plain words. Grape picks the search terms.
- 0410 ms
Search
Paragraphs ranked by how much of the question they cover.
- 051 call
Answer
Your LLM reads only those few passages.
Less work than RAG, twice.
0LLM calls for an off-topic question. The search alone refuses it.
0.1 sUntil a changed document is searchable. Vector RAG re-embeds it for 98 to 140 s.