Gemini API – Managed RAG/File Search (blog.google)

Google announced the File Search Tool for the Gemini API, a fully managed retrieval-augmented generation (RAG) system that removes the plumbing of building retrieval pipelines so developers can focus on applications. File Search handles file storage, optimal chunking, embedding creation at index time, vector search using the gemini-embedding-001 model, and dynamic injection of retrieved context into the existing generateContent API. Responses include built-in citations and support many file types (PDF, DOCX, TXT, JSON and common code files). A demo is available in Google AI Studio (requires a paid API key). Technically and economically this is significant: Google is making storage and embedding generation at query time free, so you only pay for creating embeddings when you first index files (priced at $0.15 per 1M tokens, using gemini-embedding-001). That billing model plus a managed vector search lowers operational overhead (no separate vector DB or custom chunking/lookup logic) and makes RAG more predictable and cost-effective to scale. Early users report real-time performance—Beam’s platform combines parallel searches across corpora in under two seconds—illustrating practical gains for building support bots, internal knowledge assistants, content discovery, and other grounded LLM applications.