Search documents by meaning and appearance

Type "corroded valve flange" and get the relevant text chunks, the photos that look like a corroded flange, and the photos that mean a corroded flange even when they have no useful surrounding text — in one ranked list. Open-source models throughout.

Example: finding equipment documentation

User query: "Show me turbine specification sheets and any photos of turbine parts."

Search results: 23 documents (0.34s)

1. ⭐ GE LM6000 gas turbine manual (PDF)
   📄 Technical specification (embedded text)
   📸 8 tagged images of rotor assembly, blade detail, housing
   Match score: 0.967

2. Siemens SGT-900 installation guide (PDF)
   📄 Technical specification — 142 chunks
   Match score: 0.954

3. Rolls-Royce Trent 700 component manual (PDF)
   📄 Technical reference (text + CAD diagrams)
   Match score: 0.891

What you can do with it

• Search text and images together. One API call returns ranked text chunks, visually-similar photos, and semantically-matched photos in a single result list.
• Captions are first-class text. Vision captions are embedded the same way as document text, so semantic search finds photo content even when the surrounding document has no relevant words.
• RBAC at retrieval. Results respect per-org scope catalogs; an agent only sees what its persona is permitted to see. Same dispatcher gate, no parallel access path.

Next steps

• Reference — browse prism_document_op operations (search, image_visual_search, image_text_search, get_chunks).
• Python and TypeScript SDKs — typed access to every document operation.
• Playground — try a hybrid search live against your tenant (sign-in required).