Support for gemini file search?

[johnkwaters]

I want to use gemini file search RAG pipeline in my llm in my python agent but can't find the correct syntax. What do you suggest?

`
import os
from livekit.plugins import google
from google.genai import types

def create_llm():
if os.environ.get("GOOGLE_API_KEY"):
# File search store ID usually looks like 'fileSearchStores/...'
store_id = os.environ.get("GEMINI_STORE_ID")

    return google.LLM(
        model="gemini-2.0-flash-exp", 
        # This is the correct parameter for built-in tools like RAG
        _gemini_tools=[
            types.Tool(
                file_search=types.FileSearch(
                    file_search_store_names=[store_id]
                )
            )
        ]
    )
return None

`

[davidzhao]

We've introduced a more native way to use Gemini models, please see provider tool docs for details

[davidzhao]

for file search, it'd be:

from livekit.plugins import google
from google.genai import types

agent = MyAgent(
    llm=...
    tools=[google.tools.FileSearch(file_search_store_names=["name"])]
)

[johnkwaters]

Thanks - the code compiles but what kind of prompt can I give the llm to cause it to use the file search and how can I tell if it is working?

[johnkwaters]

I have tried all kind of prompts but don't know what to refer to the tool as, so i get errors like
unknown AI function search {"function": "search", "speech_id": "speech_ba73a234a6d1"}
or unknown AI function search {"function": "filesearch", "speech_id": "speech_ba73a234a6d1"} etc

[johnkwaters]

For instance I tried this:
file_search_instructions = (
"\n\nYou have access to a file_search tool that can find relevant course materials from the document store. "
"When a student asks about course content, specific topics, assignments, or course-related questions, "
"use the file_search tool to look up relevant course materials. "
"Provide accurate answers based on the search results and cite the documents you used."
)

@davidzhao any tips?