Perception
Ground
nace ground https://example.com/report.pdf \
--target "total=1,200.50" \
--target "Jane Smith"| Flag | What it does |
|---|---|
--target [ID=]TEXT | One quote to find. Repeatable. It's split at the first =, so total=1,200.50 has the id total. Without =, the id is t<N>, where N is its place among the --target flags: t2 above. A quote that contains = needs an id: --target "rule=a=b". |
--targets JSON | Targets as a JSON array of {"id", "text", ...} objects: inline, @FILE, a path, or -. Add fields such as hint and sheet. Its entries come before any --target. |
--max-matches N | How many locations to return per target, 1 to 50, sent as options.max_matches. Default 10. |
--previews | Link a cropped image of each visual hit, by sending options.include_previews: true. |
--semantic | Sets semantic: true on every target that doesn't set semantic itself, so each target's text describes what to find instead of quoting it. On deployments that ground PDF, Word and PowerPoint sources with the exact-match runtime, the answering sentences, table cells, lines, blocks, sections or table rows come back, each with match_method semantic. semantic belongs to each target; the request has no switch of its own. |
At least one --target or --targets is required, or targets in --options. Ground reads the whole document, so it takes neither -p nor --mode. See Ground.