agent.py

import time
from core.llm_engine import LLMEngine
from core.planner import Planner
from core.executor import Executor
from core.self_improver import SelfImprover
from core.tool_registry import ToolRegistry
from core.memory import Memory  # assume your memory.py handles chat/task memory

class WindowsAgent:
    def __init__(self):
        print("[Agent] Initializing agent...")
        self.llm = LLMEngine(model="deepseek-r1:7b")
        self.planner = Planner(self.llm)
        self.tool_registry = ToolRegistry()
        self.self_improver = SelfImprover()
        self.executor = Executor(self.tool_registry, self.self_improver)
        self.memory = Memory()  # store past instructions, plans, results

    def handle_instruction(self, instruction: str):
        print(f"[Agent] Received instruction: {instruction}")

        # Save to memory
        self.memory.add_user_instruction(instruction)

        # Step 1: Generate JSON plan
        plan = self.planner.create_plan(instruction)
        print(f"[Agent] Plan generated by LLM: {plan}")

        # Step 2: Execute plan
        self.executor.execute_plan(plan)

        # Step 3: Save execution result
        self.memory.add_plan_result(plan)
        print("[Agent] Instruction executed.\n")

    def run(self):
        print("[Agent] Ready to receive instructions. Type 'exit' to quit.")
        while True:
            instruction = input("Instruction: ")
            if instruction.lower() in ("exit", "quit"):
                print("[Agent] Exiting...")
                break
            if not instruction.strip():
                continue
            try:
                self.handle_instruction(instruction)
            except Exception as e:
                print(f"[Agent] Error: {e}")
                # Optional: Try to recover or self-improve
                continue

if __name__ == "__main__":
    agent = WindowsAgent()
    agent.run()