Smart Search
Poly's search goes well beyond matching filenames. It understands the content inside your files, such as the text in a PDF, the objects in a photo, the dominant colors in an image, and uses that to find exactly what you're looking for.
What is Smart Search?
Smart Search is Poly's unified search engine based on our breakthrough search AI model, polyembed. It combines several types of search behind a single search bar:
- Full-text search — matches exact words and phrases found inside documents
- Semantic search — understands meaning, concepts, relationships, and even context, so you can find documents containing the answer to a question or look for files related to a topic, even if the exact words aren't present
- Visual similarity — find images that look like a reference image you provide
- Color search — find images that contain a specific color or palette
- Metadata search — filter by file type, size, date, and other attributes
You don't have to choose which type of search to run. Type a query and Poly figures out the most relevant results across all these dimensions.
How to run a search
Click the search pill at the top of the screen, and type your query. Results appear as you type.
By default, search covers your Home and the Home of every Shared Drive you belong to (archives are excluded). To search within one folder, right-click it and choose Search within...
What makes a file searchable
Every file runs through an intelligence pipeline after upload, and each stage unlocks a different kind of search:
| Stage | What it extracts | What it unlocks |
|---|---|---|
| Metadata | Name, type, size, dates, tags, notes | Search by name and property filters — runs for every file |
| Content | The file's raw bytes | Prerequisite for everything below |
| Similarity | Vector embeddings | Semantic search, visual similarity, related-file suggestions |
| Transcoding | Streamable video, rendered pages, archived link content | Previews, and the input for extraction below |
| Structure | Text, cell data, slide notes, transcripts, chapters | Full-text search inside the file; what the agent reads |
Two consequences worth knowing:
- A file becomes findable by name immediately, while content and semantic search light up moments later as the deeper stages complete. Writes from the CLI can block on this with
--wait-for-indexing. - A file (or folder) marked Hidden skips the pipeline entirely — it won't appear in any search and the agent can't read it. Hiding a folder cascades to everything inside.
You can watch the live status of every stage for any file in the details panel's Intelligence Features section — the first place to look when something isn't showing up in results.
What to read next
- Text queries — full-text and semantic search
- Searching by file — find similar files using one you have
- Searching by file part — search with an image region, a time slice, or a page range
- Searching by color — find images and videos by color
- Faceted search & advanced search — filter by type, date, size, and more
- Intelligence Features FAQ — troubleshoot why a file isn't appearing in search