feat(agent): implement supervisor agent with tool routing and math solver

- Replace single websearch agent with supervisor agent architecture
- Add math_solver tool using sympy for calculation queries
- Implement tool routing logic (math vs web search) in supervisor prompt
- Add interactive chat loop with conversation history
- Enhance terminal output with colored logging (colorama)
- Update dependencies: sympy, colorama

agent/agent.py

@@ -1,40 +1,53 @@
 import os
+from datetime import datetime, timezone
 from dotenv import load_dotenv
+from colorama import Fore, Style
 from langchain.agents import create_agent
-from agent.tools.search import web_search
+from langchain_core.messages import HumanMessage
+from agent.tools.math_solver import math_solver
+from agent.agents.websearch import websearch_agent
 
 load_dotenv()
 
 
-def websearch_agent(query: str) -> str:
-    """Create a simple agent with Gemini and Tavily search."""
-    base_agent = create_agent(
+def supervisor_agent():
+    """Return a supervisor agent instance with math_solver and websearch_agent."""
+    return create_agent(
         model="google_genai:gemini-3-flash-preview",
-        tools=[web_search],
+        tools=[math_solver, websearch_agent],
+        system_prompt=(
+            f"You are a supervisor agent. "
+            f"Current time is: {datetime.now(timezone.utc).isoformat()}. "
+            f"Your memory are out of date. "
+            f"For math or calculation questions, use the math_solver tool. "
+            f"For questions that need real-time, use the websearch_agent tool. "
+            f"Provide a concise and accurate final answer."
+        ),
     )
-    result = base_agent.invoke(
-        {
-            "messages": [
-                {
-                    "role": "system",
-                    "content": "Try to search web sites and get the answer.",
-                },
-                {"role": "user", "content": query},
-            ]
-        }
-    )
-    content = result["messages"][-1].content[0]
-    return content["text"]
 
 
 def run(query: str) -> str:
-    """Let Multiple agents finish the work"""
-    result = ""
-    result = websearch_agent(query)
-    return result
+    """Entry point: let the supervisor agent finish the work."""
+    print(f"{Fore.CYAN}[Supervisor] Processing query...{Style.RESET_ALL}")
+    agent = supervisor_agent()
+    result = agent.invoke({"messages": [HumanMessage(content=query)]})
+    content = result["messages"][-1].content
+    if isinstance(content, list):
+        return content[0].get("text", "")
+    return str(content)
 
 
 if __name__ == "__main__":
-    query = input("Query:\n")
-    answer = run(query)
-    print(answer)
+    agent = supervisor_agent()
+    chat_history: list = []
+    while True:
+        query = input("\nYou: ")
+        if query.lower() in ("exit", "quit"):
+            break
+        chat_history.append(HumanMessage(content=query))
+        result = agent.invoke({"messages": chat_history})
+        chat_history = result["messages"]
+        content = chat_history[-1].content
+        if isinstance(content, list):
+            content = content[0].get("text", "")
+        print(f"Agent: {content}")

agent/agents/__init__.py

@@ -0,0 +1,3 @@
+from agent.agents.websearch import websearch_agent
+
+__all__ = ["websearch_agent"]

agent/agents/websearch.py

@@ -0,0 +1,47 @@
+from datetime import datetime, timezone
+from colorama import Fore, Style  # type: ignore[import]
+from langchain_core.tools import tool
+from langchain.agents import create_agent
+from agent.tools.search import web_search
+
+
+@tool
+def websearch_agent(query: str) -> str:
+    """
+    A web search agent that searches the internet and returns an answer.
+    Use this tool when you need to find real-time or factual information
+    from the web, such as current events, specific facts, or any
+    knowledge that may require up-to-date sources.
+
+    Args:
+        query: The question or search query to look up on the web.
+    """
+    print(f"{Fore.YELLOW}[SupervisorAgent -> WebSearchAgent] {query}{Style.RESET_ALL}")
+    base_agent = create_agent(
+        model="google_genai:gemini-3-flash-preview",
+        tools=[web_search],
+        system_prompt=(
+            f"Current time is: {datetime.now(timezone.utc).isoformat()}. "
+            f"Your memory are out of date. "
+            f"All of truth that you believe without search are wrong. "
+            f"You must search the web and find the lastest answer."
+        ),
+    )
+    result = base_agent.invoke({"messages": [{"role": "user", "content": query}]})
+    content = result["messages"][-1].content
+    if isinstance(content, list):
+        content = content[0].get("text", "")
+    else:
+        content = str(content)
+    print(
+        f"{Fore.YELLOW}[WebSearchAgent -> SupervisorAgent] {content}{Style.RESET_ALL}"
+    )
+    return content
+
+
+if __name__ == "__main__":
+    from dotenv import load_dotenv
+
+    load_dotenv()
+    answer = websearch_agent.invoke({"query": "What is LangGraph?"})
+    print(answer)

agent/tools/__init__.py

@@ -1,3 +1,4 @@
 from agent.tools.search import web_search
+from agent.tools.math_solver import math_solver
 
-__all__ = ["web_search"]
+__all__ = ["web_search", "math_solver"]

agent/tools/math_solver.py

@@ -0,0 +1,47 @@
+from colorama import Fore, Style  # type: ignore[import]
+from langchain_core.tools import tool
+from sympy import sympify, N  # type: ignore[import]
+
+
+@tool
+def math_solver(expression: str) -> str:
+    """
+    Evaluate a mathematical expression using SymPy symbolic computation.
+
+    This tool converts a string expression into a SymPy symbolic object and
+    evaluates it with arbitrary precision. Unlike standard floating-point
+    arithmetic, SymPy uses exact rational and symbolic representations
+    internally, so results are free from floating-point rounding errors.
+
+    Supported syntax includes:
+    - Basic arithmetic: +, -, *, /, ** (power)
+    - Fractions: "1/3 + 2/7" (exact rational arithmetic)
+    - Square roots: "sqrt(2)", "sqrt(3)/2"
+    - Trigonometric functions: "sin(pi/4)", "cos(pi/3)", "tan(pi/6)"
+    - Logarithms: "log(100)", "ln(e**2)"
+    - Constants: "pi", "E" (Euler's number)
+    - Factorials: "factorial(10)"
+    - Combinations: "binomial(10, 3)"
+    - Complex expressions: "1/3 + 2/7 * sqrt(2)"
+
+    Args:
+        expression: A mathematical expression string to evaluate.
+                    Example: "1/3 + 2/7 * sqrt(2)"
+
+    Returns:
+        A string containing both the exact symbolic result and a numerical
+        approximation to 10 decimal places.
+    """
+    try:
+        expr = sympify(expression)
+        exact = str(expr)
+        numeric = str(N(expr, 10))
+        print(f"{Fore.MAGENTA}[MathSolver] {expression}{Style.RESET_ALL} = {numeric}")
+        return f"Exact: {exact}\nNumeric (10 digits): {numeric}"
+    except Exception as e:
+        return f"Error evaluating expression: {e}"
+
+
+if __name__ == "__main__":
+    result = math_solver.invoke({"expression": "1/3 + 2/7 * sqrt(2)"})
+    print(result)

agent/tools/search.py

@@ -1,4 +1,5 @@
 import os
+from colorama import Fore, Style
 from langchain_core.tools import tool
 from tavily import TavilyClient  # type: ignore[import]
 
@@ -14,7 +15,7 @@ def web_search(query: str, depth: str = "advanced", max_results: int = 5) -> str
         depth: Search depth, either "basic" or "advanced".
         max_results: Number of results to return (recommended 3-5 to save tokens).
     """
-    print(f"[Search & Extract] {query}")
+    print(f"{Fore.GREEN}[Search & Extract] {query}{Style.RESET_ALL}")
     client = TavilyClient(api_key=os.environ["TAVILY_API_KEY"])
 
     # 1. Search and get URLs
@@ -40,7 +41,7 @@ def web_search(query: str, depth: str = "advanced", max_results: int = 5) -> str
             item["url"]: item["raw_content"] for item in extraction.get("results", [])
         }
     except Exception as e:
-        print(f"Extraction failed: {e}")
+        print(f"{Fore.RED}Extraction failed: {e}{Style.RESET_ALL}")
         extracted_results = {}
 
     # 3. Format output

requirements.txt

@@ -7,3 +7,5 @@ langchain-core
 langchain-google-genai
 langgraph
 tavily-python
+sympy
+colorama