Searching by File

Sometimes the best description of what you're looking for is a file you already have. Poly can use any supported file as the query itself: right-click it, choose Find similar files..., and your library is ranked by similarity to it.

How it works

Every supported file gets a vector embedding when it's indexed (a representation of what it looks like, says, or sounds like). Searching by file compares the reference file's embedding against everything else you have. For images that means visual similarity: composition, palette, subject. For documents it's about textual meaning: papers can match if they contain the answer to your query, or relate to the topic, even if the exact words aren't present. Because documents are embedded per page, matches are found at page granularity. See Searching by File Part for using just a region or slice as the query.

Running one

1. Right-click a file either in the browser or on the breadcrumb path, and choose Find similar files...
2. Search opens with a Similar to chip holding the file; results rank by similarity.
3. Click the chip's thumbnail to swap in a different reference file, or × to remove it.

You can also start from the expanded search panel: Similarity → Content similarity, then pick the file.

Works with

Images, audio, PDFs, documents, presentations, spreadsheets, text files, and links. (For video content, search by a time-range selection from the player.)

From the terminal

# Everything similar to a reference file
poly search --near-file //Home/designs/logo-v3.png

# Similar PDFs only, in a folder, with a steering query
poly search "pricing tables" --near-file //Home/docs/proposal.pdf \
  --include-type pdf --under //Home/clients

The reference file itself is excluded from results by default (--ignore-files-in-query=false includes it).