Today, we're introducing the File Search Tool, a fully managed Rag program built directly on the Gemini API that releases the return pipeline so you can focus on building. File search provides a simple, integrated and critical way to return Gemini with your data, delivering accurate, relevant and verified answers.
To make file search easy and affordable for all developers, we make saving and embedding advice at query time free. You only pay to create a declaration when you start indexing your files, at a rate of $ 0.15 for 1 token 1 This new billing paradigm makes the file search tool both very simple and very expensive to build and scale as well.
How File Search Works
File search speeds up your development workflow by handling the hassle of ragging for you. It provides an easy-to-use alternative to automatic setup.
- A simple, integrated developer experience: We understand the whole process of rag. File search Automatically controls file storage, advanced chunking techniques, embedding and dynamic injection of retrievals returned to your context. It works inside the `Jegatecontent` api, which makes it easy to adopt.
- Dynamic vector search: Powered by our latest gemini-of-the-art gedini model, file search uses vector search to understand the understanding and context of the user's query. You can find relevant information from your documents, even if the exact words are not used.
- Built-in quotes: Model responses automatically include citations that describe which parts of your text are used to generate the response, making validation easier.
- Support for many different methods: You can create a complete database using a large number of file formats, including PDF, Docx, TXT, JSON and many types of languages (see the complete list of supported document formats)
You can see the file search tool in action with one of our new demo apps in Google AI Studio (requires a paid API key).
