Add tool‑use plugin: call external APIs (weather, calculators, search)

So far, your assistant can:

• Answer using its pre-trained knowledge

• Retrieve info from a local RAG knowledge base

But what if it needs live data? 👉 “What’s the weather today?” 👉 “What’s 23.5% of 450?” 👉 “Search the latest news for me.”

A purely static LLM can’t do these by itself — so you extend its capabilities by adding tool-use plugins.

---

✅ What Is Tool Use?

Tool use means:

• The assistant recognizes it can’t answer from memory.

• It calls a real external tool (API, database, function).

• It returns the result back to the user in natural language.

This makes your AI act like an agent — not just a static chatbot.

---

✅ Common Tools You Might Add

Tool	Example API
Weather	OpenWeatherMap, WeatherAPI
Calculator	Python’s eval(), or a custom math parser
Web Search	DuckDuckGo API, SerpAPI
News	NewsAPI.org
Database	SQL or vector DB queries

---

✅ How It Works

1️⃣ Detect intent — Figure out if the user’s prompt needs a tool. 2️⃣ Call the tool — Send an API request or run a Python function. 3️⃣ Combine the tool output with a natural LLM reply.

---

✅ Practical Example: Weather Plugin

---

1️⃣ Add a Python Function

import requests

def get_weather(city):
    API_KEY = "YOUR_OPENWEATHERMAP_KEY"
    url = f"http://api.openweathermap.org/data/2.5/weather?q={city}&appid={API_KEY}&units=metric"
    response = requests.get(url).json()
    temp = response['main']['temp']
    desc = response['weather'][0]['description']
    return f"The weather in {city} is {temp}°C with {desc}."

---

2️⃣ Detect When to Use It

Example: simple keyword check.

def check_tools(user_input):
    if "weather" in user_input.lower():
        return "weather"
    elif "calculate" in user_input.lower() or "%" in user_input:
        return "calculator"
    elif "search" in user_input.lower():
        return "search"
    else:
        return None

---

3️⃣ Handle the Tool Call

def handle_tools(user_input):
    tool = check_tools(user_input)

    if tool == "weather":
        # Extract city (simple example)
        city = user_input.split("in")[-1].strip()
        return get_weather(city)

    elif tool == "calculator":
        # Safe eval example
        expression = user_input.replace("calculate", "").strip()
        try:
            result = eval(expression)
            return f"The result is {result}."
        except:
            return "Sorry, I couldn’t compute that."

    elif tool == "search":
        return "Sorry, I don’t have search yet. Try asking something else."

    else:
        return None

---

4️⃣ Combine with LLM Fallback

def answer(user_input):
    tool_result = handle_tools(user_input)
    if tool_result:
        return tool_result

    # If no tool, run normal RAG + LLM
    query_embedding = embedder.encode([user_input])
    D, I = index.search(np.array(query_embedding), k=3)
    retrieved_chunks = [docs[idx] for idx in I[0]]
    context = "\n\n".join(retrieved_chunks)

    prompt = f"""Use the context to answer the question.

Context:
{context}

Question: {user_input}

Answer:"""

    output = generator(prompt, max_length=200)
    return output[0]["generated_text"]

---

✅ Now your assistant can:

• Answer normal questions using the LLM

• Call a live tool for weather or math when needed

---

✅ How to Make This Smarter

• Use intent classification with an LLM instead of keyword hacks.

• Chain tools with langchain or Autogen.

• Add logging so you know when a tool was used.

• Validate and sanitize user input for safety (don’t eval() blindly).

---

✅ When to Add Plugins

Tool Use	Example Use
Quick answers	Weather, stock prices, exchange rates
Tasks	Math, unit conversions
Data updates	News headlines
Backend integration	CRM or database lookups

---

🗝️ Key Takeaway

Tool-use = static LLM + live data = smarter assistant. It makes your chatbot truly interactive, not just a text regurgitator.

---

➡️ Next: Learn how to chain multiple tools together with an orchestration framework — or add simple error handling to keep your agent safe!