Introducing the File Search Instrument in Gemini API
Right now, we’re launching the File Search Tool, a completely managed RAG system constructed instantly into the Gemini API that abstracts away the retrieval pipeline so you’ll be able to concentrate on constructing. File Search supplies a easy, built-in and scalable technique to floor Gemini together with your knowledge, delivering responses which are extra correct, related and verifiable.
To make File Search easy and inexpensive for all builders, we’re making storage and embedding technology at question time freed from cost. You solely pay for creating embeddings if you first index your information, at a hard and fast fee of $0.15 per 1 million tokens (or regardless of the relevant embedding mannequin price is, on this case gemini-embedding-001). This new billing paradigm makes the File Search Instrument each considerably simpler and really cost-effective to construct and scale with.
How File Search works
File Search accelerates your growth workflow by dealing with the complexities of RAG for you. It supplies a user-friendly different to a self-managed setup.
- Easy, built-in developer expertise: We have streamlined your complete RAG course of. File Search robotically manages file storage, optimum chunking methods, embeddings and the dynamic injection of retrieved context into your prompts. It really works throughout the present `generateContent` API, making it simple to undertake.
- Highly effective vector search: Powered by our newest state-of-the-art Gemini Embedding model, File Search makes use of vector search to know the which means and context of a person’s question. It could possibly discover related info out of your paperwork, even when the precise phrases aren’t used.
- Constructed-in citations: The mannequin’s responses robotically embody citations that specify which components of your paperwork had been used to generate the reply, simplifying verification.
- Assist for a variety of codecs: You may construct a complete data base utilizing an enormous array of file codecs, together with PDF, DOCX, TXT, JSON and plenty of frequent programming language file sorts (see the total record of supported codecs in the docs)
You may see the File Search Instrument in motion by way of one in all our new demo app in Google AI Studio (wants a paid API key).
