Loading page content
Load a document and query it for the passages that answer your question, with everything happening in your browser: no upload, which matters when the document is a contract or an incident report.
Paste a document and ask questions about it. Uses TF-IDF keyword matching to find the most relevant paragraphs. No API calls, so all processing happens in your browser.
Retrieval over a document is only as good as its chunking. Split too small and an answer spanning two chunks is never assembled; split too large and the relevant sentence is diluted by surrounding noise. Overlapping chunks help, at the cost of duplication. This is the parameter that determines quality more than anything else in a retrieval pipeline.
Always check the retrieved passage against the source. Retrieval surfaces what is textually similar to your question, which is not the same as what answers it. A passage discussing a topic can rank highly while containing the opposite of what you assumed.