Science: Integration of Zotero #66
Description
AI Studio should have the ability to use data directly from Zotero. In this case, Zotero should be selectable as a data source, for example, in a chat. AI Studio will then vectorize (embed) all data that exists in Zotero, including metadata and file attachments such as PDFs. This will differentiate our implementation from MCP servers like Zotero MCP. Our implementation has the following advantages:
- No MCP server installation is required. AI Studio communicates directly with the Zotero database.
- By embedding the data in the latent space, we can find records that are semantically or conceptually relevant to the prompt, even if the terms or metadata do not indicate a match.
- Since our solution is based on RAG, in principle all LLMs can interact with Zotero. MCP, on the other hand, requires a model that has been trained for function calling (also known as tool use).