Function Calling Google Gemini Agent

This notebook shows you how to use Google Gemini Agent, powered by function calling capabilities.

Google's Gemini 2.5 Pro, Gemini 2.5 Flash, Gemini 2.5 Flash-Lite, Gemini 2.0 Flash models support function calling capabilities. You can find a comprehensive capabilities overview on the model overview page.

Initial Setup

Let's start by importing some simple building blocks.

The main thing we need is:

the Google Gemini API (using our own llama_index LLM class)
a place to keep conversation history
a definition for tools that our agent can use.

If you're opening this Notebook on colab, you will probably need to install LlamaIndex 🦙.

python

%pip install llama-index-llms-google-genai llama-index -q

python

# import os

# os.environ["GOOGLE_API_KEY"] = "..."

Let's define some very simple calculator tools for our agent.

python

def multiply(a: int, b: int) -> int:
    """Multiple two integers and returns the result integer"""
    return a * b


def add(a: int, b: int) -> int:
    """Add two integers and returns the result integer"""
    return a + b

Make sure your GOOGLE_API_KEY is set. Otherwise explicitly specify the api_key parameter.

python

from llama_index.llms.google_genai import GoogleGenAI
from google.genai import types

llm = GoogleGenAI(
    model="gemini-2.5-flash",
    generation_config=types.GenerateContentConfig(
        thinking_config=types.ThinkingConfig(
            thinking_budget=0
        )  # Disables thinking
    ),
)

Initialize Google Gemini Agent

Here we initialize a simple Google Gemini Agent agent with calculator functions.

python

from llama_index.core.agent.workflow import FunctionAgent

agent = FunctionAgent(
    tools=[multiply, add],
    llm=llm,
)

python

from llama_index.core.agent.workflow import ToolCallResult


async def run_agent_verbose(query: str):
    handler = agent.run(query)
    async for event in handler.stream_events():
        if isinstance(event, ToolCallResult):
            print(
                f"Called tool {event.tool_name} with args {event.tool_kwargs}\nGot result: {event.tool_output}"
            )

    return await handler

Chat

python

response = await run_agent_verbose("What is (121 + 2) * 5?")
print(str(response))

python

# inspect sources
print(response.tool_calls)

Managing Context/Memory

By default, .run() is stateless. If you want to maintain state, you can pass in a context object.

python

from llama_index.core.workflow import Context

agent = FunctionAgent(llm=llm)
ctx = Context(agent)

response = await agent.run("My name is John Doe", ctx=ctx)
response = await agent.run("What is my name?", ctx=ctx)

print(str(response))

Google Gemini Agent over RAG Pipeline

Build a Anthropic agent over a simple 10K document. We use OpenAI embeddings and Gemini 2.0 Flash to construct the RAG pipeline, and pass it to the Gemini 2.5 Flash agent as a tool.

python

!mkdir -p 'data/10k/'
!wget 'https://raw.githubusercontent.com/run-llama/llama_index/main/docs/examples/data/10k/uber_2021.pdf' -O 'data/10k/uber_2021.pdf'

python

from llama_index.core.tools import QueryEngineTool
from llama_index.core import SimpleDirectoryReader, VectorStoreIndex
from llama_index.embeddings.openai import OpenAIEmbedding
from llama_index.llms.google_genai import GoogleGenAI

embed_model = OpenAIEmbedding(model_name="text-embedding-3-large")
query_llm = GoogleGenAI(model="gemini-2.0-flash")

# load data
uber_docs = SimpleDirectoryReader(
    input_files=["./data/10k/uber_2021.pdf"]
).load_data()

# build index
uber_index = VectorStoreIndex.from_documents(
    uber_docs, embed_model=embed_model
)
uber_engine = uber_index.as_query_engine(similarity_top_k=3, llm=query_llm)
query_engine_tool = QueryEngineTool.from_defaults(
    query_engine=uber_engine,
    name="uber_10k",
    description=(
        "Provides information about Uber financials for year 2021. "
        "Use a detailed plain text question as input to the tool."
    ),
)

python

from llama_index.core.agent.workflow import FunctionAgent

agent = FunctionAgent(tools=[query_engine_tool], llm=llm, verbose=True)

python

response = await agent.run(
    "Tell me both the risk factors and tailwinds for Uber?"
)
print(str(response))