Google has expanded its Gemini API File Search tool, now supporting multimodal data for retrieval-augmented generation systems. Developers can now build applications that process both text and images natively, powered by the Gemini Embedding 2 model. This enhancement provides AI agents with contextual awareness beyond text-only processing.

The update introduces custom metadata filtering, allowing users to attach key-value labels to unstructured data. Metadata tags like "department: Legal" or "status: Final" enable precise data slicing at query time, reducing noise from irrelevant documents. This improves both the speed and accuracy of information retrieval in large document repositories.

Page citations now tie model responses directly to original source documents, capturing page numbers for every piece of indexed information. This granularity builds user trust and enables rigorous fact-checking. Applications can point users directly to the source material, making the tool immediately useful for verification workflows in professional environments.

These enhancements address common challenges in document search and retrieval. Creative agencies can search visual archives by emotional tone rather than filenames, while legal professionals filter documents by custom attributes. The tool handles infrastructure complexity, letting developers focus on building products with more reliable, verifiable AI capabilities.