Gemini’s New File Search Just Leveled Up RAG Agents (10x Cheaper)

This video demonstrates using Gemini File Search API to build RAG agents in Naden. It shows a practical workflow: create a file store, upload documents, import them into the store, and query with a Gemini-backed agent. The presenter compares costs to common vector stores, runs an accuracy evaluation across multiple PDFs, and discusses real-world limits like duplication, chunking, and privacy.

– Workflow steps: create a store (folder), upload files via a resumable upload, import files into the store, then call the generate/content endpoint to query via an AI agent.

– Pricing and tradeoffs: indexing costs are low (15¢ per 1M tokens), storage is currently free on Gemini, and querying costs are mainly model usage; compared with Pinecone/Supabase, Gemini simplifies setup while remaining inexpensive.

– Demo & evaluation: tested three PDFs (Apple 10-K, NVIDIA release, golf rules) with 10 questions and achieved ~4.2/5 correctness overall.

– Practical caveats: not magic — watch for duplicate/updated files, OCR or messy inputs, chunk-based retrieval limits for whole-document queries, and sensitive data / compliance concerns (GDPR, HIPAA).

Quotes:

You basically just upload the doc, we shoot it over to Google, and then it takes care of everything else.

They only charge you for uploading or for your indexing or your embedding — 15 cents for every 1 million tokens processed.

This isn’t magic… garbage in is garbage out.