Searching by File Part
A whole file is often too much query. The logo is in the corner of the screenshot; the quote is two minutes of an hour-long recording; the diagram is one page of a forty-page spec. Poly can search by just the part that matters.
This works because Poly indexes files at part granularity — documents are embedded page by page, audio and video in time segments — so a partial query matches against parts, and results point to the matching page or moment, not just the file.
Image regions
Drag a rectangle over any image in the preview to select a region, then right-click the selection and choose Find similar files. Only what's inside the box drives the search — crop to the product, the logo, the face of the building, and matches rank by similarity to that region alone.
Audio & video slices
In the audio or video player, drag a selection across the timeline, right-click it, and choose Find similar files. The selected slice — what's said and, for video, what's shown in that window — becomes the query. Perfect for "find the other takes of this section" or "where else does this topic come up?".
While a time selection is active you can also choose Search inside this file to match against the rest of the same recording instead of your whole library.
Document pages and text spans
From the terminal, the CLI can anchor a search to any slice of any file, including two kinds the app doesn't expose yet, page ranges and line ranges:
# Match against pages 3–5 of a PDF
poly search --near-file //Home/specs/design.pdf --near-file-pages "3-5"
# Match against two minutes of a recording
poly search --near-file //Home/media/standup.mp4 \
--near-file-start-time "12:40" --near-file-end-time "14:40"
# Match against a span of a text file
poly search --near-file //Home/notes/draft.md \
--near-file-start-line 40 --near-file-end-line 80
You can only specify one slice type per search using --near-file. A steering query and any filters combine with the slice just like a whole-file search.
Results land on the matching part
Because matching happens per part, opening a result takes you to the right place: the matching page of a PDF, the matching moment of a recording. The same mechanics power text-query matches, too.