tools/math_agent.py

from typing import Literal

from langchain.chat_models import init_chat_model
from langchain.tools import tool
from langchain_core.messages import SystemMessage, ToolMessage, filter_messages, HumanMessage
from langgraph.graph import StateGraph, START, END

from core.state import MathAgentState, MathAgentOutputState
from tools.python_executor import execute_python_code
from tools.think_tool import think_tool
from utils.prompt_manager import prompt_mgmt

# ===== CONFIGURATION =====

# Set up tools and model binding
tools = [execute_python_code, think_tool]
tools_by_name = {tool.name: tool for tool in tools}

# Initialize models
math_model = init_chat_model(model="openai:gpt-5")
model_with_tools = math_model.bind_tools(tools)
summarization_model = init_chat_model(model="openai:gpt-4.1-mini")
compress_model = init_chat_model(model="openai:gpt-4.1",
                                 max_tokens=32000)


def llm_call(state: MathAgentState):
    """Analyze current state and decide on next actions.

    The model analyzes the current state and decides whether to:
    1. Call search tools to gather more information
    2. Provide a final answer based on gathered information

    Returns updated state with the model's response.
    """
    return {
        "messages": [
            model_with_tools.invoke(
                [SystemMessage(content=prompt_mgmt.render_template("math_agent_base_system", {}))] + state[
                    "messages"]
            )
        ]
    }


def tool_node(state: MathAgentState):
    """Execute all tool calls from the previous LLM response.

    Executes all tool calls from the previous LLM responses.
    Returns updated state with tool execution results.
    """
    tool_calls = state["messages"][-1].tool_calls

    # Execute all tool calls
    observations = []
    for tool_call in tool_calls:
        tool = tools_by_name[tool_call["name"]]
        observations.append(tool.invoke(tool_call["args"]))

    # Create tool message outputs
    tool_outputs = [
        ToolMessage(
            content=observation,
            name=tool_call["name"],
            tool_call_id=tool_call["id"]
        ) for observation, tool_call in zip(observations, tool_calls)
    ]

    return {"messages": tool_outputs}


def compress_research(state: MathAgentState) -> dict:
    """Compress research findings into a concise summary.

    Takes all the research messages and tool outputs and creates
    a compressed summary suitable for the supervisor's decision-making.
    """

    last_message = state.get("messages", [])[-1]

    # Extract raw notes from tool and AI messages
    raw_notes = [
        str(m.content) for m in filter_messages(
            state["messages"],
            include_types=["tool", "ai"]
        )
    ]

    return {
        "compressed_research": str(last_message.content),
        "raw_notes": ["\n".join(raw_notes)]
    }


def should_continue(state: MathAgentState) -> Literal["tool_node", "compress_research"]:
    """Determine whether to continue the solving of the problem or provide final answer.

    Determines whether the agent should continue the solving loop or provide
    a final answer based on whether the LLM made tool calls.

    Returns:
        "tool_node": Continue to tool execution
        "compress_research": Stop and compress research
    """
    messages = state["messages"]
    last_message = messages[-1]

    # If the LLM makes a tool call, continue to tool execution
    if last_message.tool_calls:
        return "tool_node"
    # Otherwise, we have a final answer
    return "compress_research"


# ===== GRAPH CONSTRUCTION =====

# Build the agent workflow
agent_builder = StateGraph(MathAgentState, output_schema=MathAgentOutputState)

# Add nodes to the graph
agent_builder.add_node("llm_call", llm_call)
agent_builder.add_node("tool_node", tool_node)
agent_builder.add_node("compress_research", compress_research)

# Add edges to connect nodes
agent_builder.add_edge(START, "llm_call")
agent_builder.add_conditional_edges(
    "llm_call",
    should_continue,
    {
        "tool_node": "tool_node",  # Continue research loop
        "compress_research": "compress_research",  # Provide final answer
    },
)
agent_builder.add_edge("tool_node", "llm_call")  # Loop back for more research
agent_builder.add_edge("compress_research", END)

# Compile the agent
math_agent = agent_builder.compile()


@tool
def math_tool(problem: str):
    """
    Tool for solving a mathematical problem
    :param problem: The problem to be solved
    :return: the solution to the given problem
    """
    response = math_agent.invoke({"messages": [HumanMessage(content=problem)], "question": problem})
    # return the LLM response as a string (expected tool response format)
    # this will be automatically turned to ToolMessage
    # by the prebuilt create_react_agent (supervisor)
    return response['compressed_research']