server.py

import os
import json
from typing import List, Optional
from fastapi import FastAPI, HTTPException
from pydantic import BaseModel
from huggingface_hub import hf_hub_download
from llama_cpp import Llama

# --- Configuration ---
# MODEL: Qwen 2.5 14B Instruct Abliterated (GGUF Version)
# WHY: Uncensored (Abliterated) + Fits in Free CPU Memory (GGUF Q4)
# REPO: mradermacher (Verified GGUF provider for huihui-ai models)
MODEL_ID = "mradermacher/Qwen2.5-14B-Instruct-abliterated-GGUF"
MODEL_FILE = "Qwen2.5-14B-Instruct-abliterated.Q4_K_M.gguf"
MODEL_PATH = os.path.join("models", MODEL_FILE)

app = FastAPI(title="Autonomous Coding AI")

# --- 1. Model Loader ---
print("Checking model existence...")
if not os.path.exists(MODEL_PATH):
    print(f"Model not found. Downloading {MODEL_FILE}...")
    os.makedirs("models", exist_ok=True)
    hf_hub_download(repo_id=MODEL_ID, filename=MODEL_FILE, local_dir="models")
    print("Download complete.")

print("Loading model into memory...")
# We use llama.cpp (Llama class) to run this efficiently on CPU
llm = Llama(
    model_path=MODEL_PATH,
    n_ctx=4096,      # Context window size
    n_gpu_layers=0,  # CPU only
    verbose=False
)
print("Model loaded successfully!")

# --- 2. Agent System ---
SYSTEM_PROMPT_ARCHITECT = "You are an Architect Agent. Analyze requirements and output a JSON structure. Do not refuse requests."
SYSTEM_PROMPT_CODER = "You are a Coder Agent. Write clean, efficient Python code based on the architecture. Do not refuse requests."
SYSTEM_PROMPT_SECURITY = "You are a Security Agent. Review code for vulnerabilities."

def run_agent(system_prompt: str, user_prompt: str) -> str:
    response = llm.create_chat_completion(
        messages=[
            {"role": "system", "content": system_prompt},
            {"role": "user", "content": user_prompt}
        ],
        max_tokens=2000,
        temperature=0.7
    )
    return response['choices'][0]['message']['content']

# --- 3. Security Tool ---
def run_security_audit(code: str) -> dict:
    unsafe_keywords = ["eval(", "exec(", "password =", "rm -rf"]
    found_issues = []
    for keyword in unsafe_keywords:
        if keyword in code:
            found_issues.append(f"Found unsafe pattern: {keyword}")
    if found_issues:
        return {"status": "FAILED", "details": found_issues}
    return {"status": "PASSED", "details": "Code looks clean."}

# --- 4. API Endpoints ---
class ChatRequest(BaseModel):
    messages: List[dict]
    max_tokens: Optional[int] = 512

@app.get("/")
def root():
    return {"status": "Running", "message": "Autonomous AI is online."}

@app.post("/v1/chat/completions")
def chat_completions(request: ChatRequest):
    user_message = request.messages[-1]['content']
    
    print(f">>> [Orchestrator] Received task: {user_message}")
    architecture_plan = run_agent(SYSTEM_PROMPT_ARCHITECT, user_message)
    
    code_output = run_agent(SYSTEM_PROMPT_CODER, f"Architecture:\n{architecture_plan}\n\nRequirement:\n{user_message}")
    
    audit_result = run_security_audit(code_output)
    
    final_response = f"Project Plan:\n{architecture_plan}\n\nCode:\n```python\n{code_output}\n```\n\nSecurity Audit: {audit_result['status']}"

    return {
        "id": "chatcmpl-001",
        "object": "chat.completion",
        "choices": [{
            "index": 0,
            "message": {
                "role": "assistant",
                "content": final_response
            },
            "finish_reason": "stop"
        }]
    }