Documentation Index
Fetch the complete documentation index at: https://mintlify.com/derailed-dash/gemini-file-search-demo/llms.txt
Use this file to discover all available pages before exploring further.
Before adding RAG, it’s worth understanding what a Gemini agent looks like without it. This page walks through app/sdk_agent.py — a minimal agent built directly on the google-genai SDK that attaches the native Google Search tool to a Gemini chat session. Running this agent first establishes a clear baseline: it handles general web-searchable questions well, but it has no knowledge of your custom documents, so questions about bespoke content will either fail or hallucinate.
The code
"""
Native SDK Basic Agent (Google Search Only)
This script demonstrates a minimal "Agent" using the raw Google Gen AI SDK.
WHAT:
A simple CLI chatbot that uses Gemini 2.5 Flash and the native Google Search tool.
WHY:
To establish a baseline for "native" (non-framework) implementation. It highlights
how easy it is to attach the `GoogleSearch` tool to a model.
HOW:
1. Instantiates `genai.Client()`.
2. Creates a chat session with `config=types.GenerateContentConfig(tools=[...])`.
3. Uses the built-in `types.GoogleSearch()` tool.
4. Runs a simple REPL loop for user interaction.
"""
import logging
import os
from dotenv import load_dotenv
from google import genai
from google.genai import types
load_dotenv()
# Configure logging
logging.basicConfig(level=logging.INFO, format="%(asctime)s - %(levelname)s - %(message)s")
logger = logging.getLogger(__name__)
def main():
"""
Runs a simple interactive chat with Gemini using the Google Search tool.
"""
# 1. Initialize the client
client = genai.Client()
# 2. Add Google Search tool
google_search_tool = types.Tool(google_search=types.GoogleSearch())
# 3. Create the chat session
model_id = os.getenv("MODEL", "gemini-2.5-flash")
logger.info(f"Starting chat with model: {model_id}")
chat = client.chats.create(
model=model_id,
config=types.GenerateContentConfig(
tools=[google_search_tool],
temperature=0.7, # slightly creative but grounded
),
)
print(f"--- Simple Agent (Model: {model_id}) ---")
print("Type 'exit' or 'quit' to stop.")
print("------------------------------------------")
while True:
try:
user_input = input("User: ").strip()
if user_input.lower() in ("exit", "quit"):
print("Goodbye!")
break
if not user_input:
continue
# 4. Send message and get response
response = chat.send_message(user_input)
# 5. Print response
if response.text:
print(f"Agent: {response.text}")
else:
print("Agent: [No text response]")
except Exception as e:
logger.error(f"An error occurred: {e}")
print(f"Error: {e}")
if __name__ == "__main__":
main()
Code walkthrough
Each of the four steps in main() has a distinct job.
Step 1 — Initialize the client
genai.Client() reads GEMINI_API_KEY from the environment (loaded via load_dotenv()) and creates the top-level client you use to access all Gemini API resources — models, chats, and later, file search stores.
google_search_tool = types.Tool(google_search=types.GoogleSearch())
types.GoogleSearch() is the built-in grounding tool that allows Gemini to issue live web searches when it needs up-to-date information. Wrapping it in types.Tool(...) produces the object that the chat session config accepts.
Step 3 — Create the chat session
chat = client.chats.create(
model=model_id,
config=types.GenerateContentConfig(
tools=[google_search_tool],
temperature=0.7,
),
)
client.chats.create(...) opens a stateful multi-turn conversation. Passing tools=[google_search_tool] inside GenerateContentConfig makes Google Search available for every turn of the conversation. The MODEL environment variable controls which Gemini model is used; it falls back to gemini-2.5-flash if unset.
Step 4 — REPL loop
while True:
user_input = input("User: ").strip()
if user_input.lower() in ("exit", "quit"):
print("Goodbye!")
break
if not user_input:
continue
response = chat.send_message(user_input)
if response.text:
print(f"Agent: {response.text}")
else:
print("Agent: [No text response]")
chat.send_message(user_input) sends the user’s message and receives the model’s reply. Because the session is stateful, Gemini maintains the full conversation history across turns.
Running the agent
Demo walkthrough
Ask a web-searchable question
Try a question Gemini can answer via Google Search:> What is the stock price of Google?
Ask about bespoke content
Now ask a question that requires knowledge of your custom story — content Gemini has never been trained on and cannot find via web search:> Who pilots the 'Too Many Pies' ship?
Exit the agent
Type quit or exit to close the session. Key takeaway
Google Search makes this agent great at general, web-indexed knowledge. It has no mechanism to access your private or bespoke documents — that’s exactly the gap Gemini File Search fills in the next steps.
