3.1 Introduction
Let’s create a simple AI Agent that will deside which tools to use based on the question. We have two tools or functioned defined -
- Calculator Tools
- Wikipedia Search tools
We will build a simple Agent that:
- Reads your question.
- Decides if it needs a calculator ot just reasoning
- calls the right tool
- Give us the answer.
First Starting with creating tools
3.2 Defining Tools
Importing ollama and requests packages
First Defining a tool that evaluate Mathematical expression. Lets call it a Calculator tool.
# 1. Simple calculator tool
def calculator_tool(query):
"""A simple calculator tool."""
try:
# Keep only numbers, operators, parentheses, and dots
cleaned_expr = query.split('=')[0] # Remove anything after '='
cleaned_expr = "".join(cleaned_expr).strip()
return str(eval(cleaned_expr))
except Exception as e:
return f"Error in calculation: {e}"Our second tool is wikipedia search tool, which extract data from wikipedia based on user question
# 2. Wikipedia search tool
def web_search(query):
url = "https://en.wikipedia.org/w/api.php"
params = {
"action": "query",
"list": "search",
"srsearch": query,
"format": "json"
}
try:
response = requests.get(url, params=params)
data = response.json()
if "query" in data and "search" in data["query"]:
results = [item['snippet'] for item in data['query']['search']]
return "\n".join(results) if results else "No results found on Wikipedia."
else:
return "No results found."
except Exception as e:
return f"Error searching Wikipedia: {e}"3.3 Defining a Simple Agent
This is a very simple use case of Agent. This Problem can also be solved without creating an Agent as well.
Our probelm statment is to create an agent that deside which tool to use based on the question. For example, if the user ask Who is the president of the USA?, the agent will suggest to use wikipedia_search.
Let’s build the Agent and see
def agent(query, model="llama3.2:latest"):
# Step 1: Ask the LLM to deside what to do
system_prompt = """
You are a helpful AI assistant. You can either perform calculations or search the web.
- CALCULATE: <math expression>
- SEARCH: <search query>
- ANSWER: <your answer>
Output ONLY one of the above actions on the FIRST line.
Do not explain your reasoning. Do not add extra commentary.
Example:
Q: What is 12*(5+2)?
A: CALCULATE: 12*(5+2)
Q: Who is the president of the USA?
A: SEARCH: Who is the president of the USA?
Q: Tell me a joke about Cats.
A: ANSWER: Sure, here is a joke:
"""
decision_response = ollama.chat(
model=model,
messages=[
{"role": "system", "content": system_prompt},
{"role": "user", "content": query}
]
)
decision = decision_response['message']['content'].strip()
# Normalize the decision
first_line = decision.split('\n')[0].strip()
# print(f"\tDecision made by LLM: {first_line}")
# STEP 2: Execute the decision
if "CALCULATE" in first_line:
expression = first_line.split("CALCULATE:", 1)[1].strip()
return calculator_tool(expression)
elif "SEARCH:" in first_line:
query = first_line.split("SEARCH:")[1].strip()
return web_search(query)
elif "ANSWER:" in first_line:
return first_line.split("ANSWER:")[1].strip()
else:
return "I didn't understand the instruction. Please try again with a clear command."Note: We are using llama3.2:latest model for the agent. Based on the model, we might need to change the prompt and how we are extracting the tool name.
3.4 Calling Agent
Let’s call the Agent and see if this is working fine or not. First try to invoke calculator.
Output is
Next try to invoke wikipedia_search
The output is
India, officially the Republic of India, is a country in South Asia. It is the seventh-largest country by area; the most populous country since 2023;
Developmental Inclusive Alliance (INDIA) is a big tent multi-party political alliance of several political parties in India led by the country's largest opposition
The economy of India is a developing mixed economy with a notable public sector in strategic sectors. It is the world's fourth-largest economy by nominal
The Constitution of India is the supreme legal document of India, and the longest written national constitution in the world. The document lays down the
India, colloquially called Tiraṅgā (the tricolour), is a horizontal rectangular tricolour flag, the colours being of India saffron, white and India green;
India is the most populous country in the world, with one-sixth of the world's population. Between 1975 and 2010, the population doubled to 1.2 billion
The Times of India (TOI) is an Indian English-language daily newspaper and digital news media owned and managed by the Times Group. It is the third-largest
Languages of India belong to several language families, the major ones being the Indo-Aryan languages spoken by 78.05% of Indians and the Dravidian languages
The president of India (ISO: Bhārata kē Rāṣṭrapati) is the head of state of the Republic of India. The president is the nominal head of the executive,
between 2500 BCE and 1900 BCE in present-day Pakistan and north-western India. Early in the second millennium BCE, persistent drought caused the populationWhat if we are asking something that does not requir any tools. In that case based on the if condition LLM answer from its memory. For example,
The Output in this case is
3.5 How this works
The Agent works as follows:
- The LLM acts as the planner.
- It decides which tool to call based on your question.
- Tools (calculator, web search) are normal Python functions.
- You can add more tools (database query, API calls, etc.).
3.6 Why This Is Agentic AI
This is not just a chatbot - it’s:
- Reasoning (LLM deciding the next step)
- Acting (calling a tool)
- Observing (getting results back)
Out of all 4 key parts of an Agent we have mentioned in last chapter this Agent has LLM and Tools.
We will be adding Memory and Looping in subsequent chapters.