Google expanded the Gemini API File Search tool to support multimodal RAG with image plus text retrieval, custom metadata filters, and page-level citations.
Key Takeaways
File Search now indexes images natively alongside text using the Gemini Embedding 2 model, enabling natural language queries over visual archives.
Custom metadata key-value labels (e.g. department: Legal, status: Final) let you scope queries to specific data slices, reducing retrieval noise.
Page citations tie model responses to exact source page numbers, enabling verifiable fact-checking in PDF-heavy workflows.
Google positions File Search as managed RAG infrastructure, handling chunking, embedding, and retrieval so developers skip that stack.
