agent_integration.py

from typing import Optional, List
from langchain.agents import Tool, AgentType, initialize_agent
from langchain.llms.base import LLM

from firedetection import detect_fire_in_video

# Define a LangChain tool using a decorator or by manually creating a Tool object.
@Tool(
    name="FireDetectionTool",
    description="Detect fire in a given video using YOLO."
)
def detect_fire_in_video_tool(video_path: str) -> str:
    """
    Takes a video path, calls 'detect_fire_in_video' function,
    and returns a message indicating the output location.
    """
    output_video_path = detect_fire_in_video(
        input_video_path=video_path,
        output_video_path="temp_outputs/result.avi",
        model_path="/Users/bahakizil/Desktop/firedetection/firedetectionyolo/last.pt",  # Adjust if your model file is named differently
        device="cpu"               # or 'cuda'/'mps' if available
    )
    return f"Fire detection completed. Output saved to: {output_video_path}"


class SimpleFakeLLM(LLM):
    """
    A trivial/fake LLM for demonstration. 
    It always decides to call 'FireDetectionTool' regardless of the prompt.
    This replaces the need for external APIs like OpenAI or HuggingFace,
    while still demonstrating the agent + tool usage in LangChain.
    """

    @property
    def _llm_type(self) -> str:
        return "fake-llm"

    def _call(self, prompt: str, stop: Optional[List[str]] = None) -> str:
        # Always produce a response suggesting the use of FireDetectionTool:
        return "I should use FireDetectionTool to process the video."

def create_fire_detection_agent():
    """
    Creates a LangChain agent with:
      - A single tool: FireDetectionTool
      - A trivial LLM (SimpleFakeLLM)
    """
    llm = SimpleFakeLLM()
    tools = [detect_fire_in_video_tool]

    agent = initialize_agent(
        tools=tools,
        llm=llm,
        agent=AgentType.ZERO_SHOT_REACT_DESCRIPTION,
        verbose=True
    )
    return agent

if __name__ == "__main__":
    # Example usage:
    agent = create_fire_detection_agent()

    # Prompt the agent to detect fire in a given video (sample_video.mp4)
    prompt = "Please detect fire in 'sample_video.mp4'."
    response = agent.run(prompt)
    print("\nAgent response:", response)