app.py

import os
import httpx
import json
import chromadb
from contextlib import asynccontextmanager
from fastapi import FastAPI, Request
from fastapi.responses import StreamingResponse
from datasets import load_dataset
from huggingface_hub import HfApi
from tavily import TavilyClient
from fastapi.middleware.cors import CORSMiddleware

# --- Secrets & Config ---
GROQ_API_KEY = os.getenv("GROQ_API_KEY")
TAVILY_API_KEY = os.getenv("TAVILY_API_KEY")
HF_TOKEN = os.getenv("HF_TOKEN") 
DATASET_ID = "AIDev07/AIModelsLoaded"
DB_PATH = "./chroma_db"

# --- Vector DB Initialization ---
chroma_client = chromadb.PersistentClient(path=DB_PATH)
collection = chroma_client.get_or_create_collection(name="project_memory")

# --- Lifespan Handler (Replaces @app.on_event) ---
@asynccontextmanager
async def lifespan(app: FastAPI):
    # Startup Logic: Sync Vector Memory
    print("🔄 Lifespan Startup: Syncing Memory...")
    if collection.count() == 0:
        try:
            ds = load_dataset(DATASET_ID, token=HF_TOKEN, split="train")
            for i, record in enumerate(ds):
                text = record.get("content") or record.get("text") or str(record)
                collection.add(documents=[text], ids=[f"doc_{i}"])
            print("✅ Vector Memory Loaded.")
        except Exception as e:
            print(f"❌ Pull Failed: {e}")
    yield
    # Shutdown Logic (Optional: Clear cache or close DB connections)
    print("🛑 Lifespan Shutdown: Cleaning up...")

app = FastAPI(lifespan=lifespan)

# --- Cross Origin Resource Sharing ---
app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],
    allow_methods=["*"],
    allow_headers=["*"],
)

# --- Helper: Save to HF ---
def save_to_hf(new_content: str):
    api = HfApi()
    try:
        filename = f"updates/update_{collection.count()}.txt"
        with open("update.txt", "w") as f:
            f.write(new_content)
        
        api.upload_file(
            path_or_fileobj="update.txt",
            path_in_repo=filename,
            repo_id=DATASET_ID,
            repo_type="dataset",
            token=HF_TOKEN
        )
        collection.add(documents=[new_content], ids=[f"doc_{collection.count()}"])
        return True
    except Exception as e:
        print(f"❌ Write Failed: {e}")
        return False

# --- Chat Endpoint (Httpx Proxy) ---
@app.post("/v1/chat")
async def chat_endpoint(request: Request):
    data = await request.json()
    messages = data.get("messages", [])
    user_query = messages[-1]["content"]

    # 1. RAG Retrieval
    docs = collection.query(query_texts=[user_query], n_results=5)
    context_str = "\n".join(docs['documents'][0]) if docs['documents'] else ""

    # 2. Dynamic Model Routing (Bypass Logic)
    model = "llama-3.3-70b-versatile"
    if any(k in user_query.lower() for k in ["code", "python", "script"]):
        model = "openai/gpt-oss-120b"
    elif "kimi" in user_query.lower():
        model = "kimi-k2-instruct-0905"

    async def stream_logic():
        headers = {"Authorization": f"Bearer {GROQ_API_KEY}"}
        # Prepend the vector context to the first system message
        payload = {
            "model": model,
            "messages": [{"role": "system", "content": f"Project Context: {context_str}"}] + messages[-5:],
            "stream": True
        }
        async with httpx.AsyncClient(timeout=120.0) as client:
            async with client.stream("POST", "https://api.groq.com/openai/v1/chat/completions", headers=headers, json=payload) as response:
                async for line in response.aiter_lines():
                    if line.startswith("data: "):
                        yield f"{line}\n\n"

    return StreamingResponse(stream_logic(), media_type="text/event-stream")

@app.post("/v1/debug")
async def debug_endpoint(request: Request):
    data = await request.json()
    messages = data.get("messages", [])
    
    headers = {"Authorization": f"Bearer {GROQ_API_KEY}"}
    payload = {
        "model": "llama-3.3-70b-versatile",
        "messages": messages,
        "stream": False  # non-streaming first to isolate the issue
    }
    
    async with httpx.AsyncClient(timeout=30.0) as client:
        response = await client.post(
            "https://api.groq.com/openai/v1/chat/completions",
            headers=headers,
            json=payload
        )
        return {
            "groq_status": response.status_code,
            "groq_key_set": bool(GROQ_API_KEY),
            "groq_key_prefix": GROQ_API_KEY[:8] if GROQ_API_KEY else "MISSING",
            "groq_response": response.json()
        }
        
# --- Floating Terminal / Shell Control ---
@app.post("/v1/shell")
async def shell_exec(request: Request):
    data = await request.json()
    command = data.get("command", "")
    
    # Internal 'save' command to trigger the HF Write bypass
    if command.startswith("save "):
        content = command.replace("save ", "")
        status = save_to_hf(content)
        return {"status": "success" if status else "failed", "action": "HF_WRITE"}
    
    return {"status": "ready_to_copy", "command": command}

if __name__ == "__main__":
    import uvicorn
    uvicorn.run(app, host="0.0.0.0", port=7860)