Gemini
Google DeepMind Launches File Search Tool for Gemini API
The new File Search Tool adds a fully managed retrieval-augmented generation (RAG) system directly into the Gemini API, giving developers an integrated and scalable way to ground responses in their own data without building a custom retrieval pipeline.
Georg S. Kuklick
•
November 6, 2025
Google DeepMind has introduced the File Search Tool, a managed RAG solution built into the Gemini API that simplifies how developers connect their data to Gemini models. The system automates key retrieval tasks such as file storage, chunking, embedding generation, and context injection, allowing teams to focus on application logic rather than data infrastructure.
The File Search Tool uses Google’s latest Gemini Embedding model to perform vector searches that capture the meaning of queries, not just their keywords. Responses include built-in citations linking back to source documents, improving transparency and traceability. Developers can use a broad set of supported file formats including PDF, DOCX, TXT, JSON, and many programming language file types.
Google has introduced a cost model designed to make RAG development more accessible. Embedding generation at query time and file storage are free, while initial embedding creation is priced at $0.15 per million tokens. The approach reduces ongoing costs and simplifies scaling for both small and enterprise deployments.
Early adopters have reported strong performance results. Phaser Studio’s Beam platform, which builds AI-driven games, uses File Search to query thousands of documents daily across multiple corpora. The company says search operations now complete in under two seconds, replacing manual cross-referencing that previously took hours.
Developers can start using the File Search Tool immediately through the Gemini API and Google AI Studio, where a demo app showcases practical implementations for knowledge assistants, support bots, and creative discovery systems.
