demeter

import os
import json
from langchain.tools import tool
from qdrant_client import QdrantClient

# Import your existing modules
from agent.sub_agents.Researcher import ResearcherAgent
from agent.Qdrant.Store import COLLECTION_NAME

from agent.sub_agents.Doctor import VisionAgent
from agent.memory import FarmMemory

# Initialize shared clients
# Note: We rely on the existing ResearcherAgent logic for embeddings/search
researcher_instance = ResearcherAgent()
farm_memory = FarmMemory()

# Initialize Qdrant for the Historian
qdrant_client = QdrantClient(
    url=os.environ.get("QDRANT_URL", "http://localhost:6333"),
    api_key=os.environ.get("QDRANT_API_KEY"),
)

doctor = VisionAgent()

@tool
def diagnose_plant(image_b64: str = None):
    """
    Uses Computer Vision to scan the latest plant image for disease.
    
    INSTRUCTIONS FOR LLM: 
    Call this tool with NO arguments. The system will automatically 
    attach the latest camera feed for you.
    """
    if not image_b64:
        return {"error": "System Error: No visual data was injected into the tool call."}
        
    # Calls the analyze_frame method you updated earlier
    result = doctor.analyze_frame(image_b64)
    print(f"Diagnose Plant Result: {result}")
    return result

@tool
def ask_memory(plant_id: str):
    """
    Consult the Farm Memory for past events, strategies, and outcomes.
    Useful for recalling what has been tried before and its results.
    
    Args:
        plant_id: The specific crop ID to look up (e.g. "crop_beta").
    """
    print(f"Consulting Farm Memory for plant ID: {plant_id}")
    try:
        # Fixed: calling the wrapper method directly on the instance
        response = farm_memory.get_plant_history(plant_id)
        print(f"Memory response for {plant_id}: {response}")
        return response
    except Exception as e:
        return f"Memory unavailable: {str(e)}"

@tool
def ask_historian(query: str):
    """
    Search the Farm's Database (History) for similar past events.
    Useful for checking past mistakes, success rates, or specific scenarios.
    
    Args:
        query: A description of the situation to look up (e.g. "What happened when pH dropped to 5.5?")
    """
    try:
        # 1. Generate Embedding using Researcher's Encoder (384 dims)
        # .embed() returns a generator, convert to list and take first item
        embeddings = list(researcher_instance.encoder.embed([query]))
        query_vector = embeddings[0].tolist()
        
        # 2. Search the TEXT collection (Plant_Biographies_HF)
        # We CANNOT search COLLECTION_NAME (Farm_Memory) because the vector dimensions don't match.
        hits = qdrant_client.query_points(
            collection_name="Plant_Biographies_HF  ",
            query=query_vector,
            limit=3
        )
        
        results = []
        for hit in hits:
            # Handle different payload structures
            content = hit.payload.get('text', hit.payload.get('memory', ''))
            source = hit.payload.get('source', hit.payload.get('user_id', 'Unknown'))
            results.append(f"[{source}]: {content}")
            
        return "\n".join(results) if results else "No relevant history found."
        
    except Exception as e:
        return f"Historian unavailable: {str(e)}"

@tool
def ask_rag(query: str):
    """
    Consult the Research Assistant (RAG) for scientific knowledge.
    Useful for finding optimal ranges, chemical interactions, or biological facts.
    
    Args:
        query: Scientific question (e.g. "Optimal VPD for Lettuce in late flower")
    """
    try:
        # Reuse your existing ResearcherAgent logic
        return researcher_instance.search(query)
    except Exception as e:
        return f"Research unavailable: {str(e)}"