app.py

import os
import gradio as gr
from openai import OpenAI
from collections import defaultdict
import time

# Loads keys from .env file when running locally
# On HuggingFace/Colab the env vars are already set so this just does nothing
try:
    from dotenv import load_dotenv
    load_dotenv()
except ImportError:
    pass  # dotenv not installed (HF/Colab don't need it)

# ── Read keys from environment variables (never hardcode these!) ──
API_KEYS = {
    "groq":        os.environ.get("GROQ_API_KEY"),
    "gemini":      os.environ.get("GEMINI_API_KEY"),
    "mistral":     os.environ.get("MISTRAL_API_KEY"),
    "cohere":      os.environ.get("COHERE_API_KEY"),
    "huggingface": os.environ.get("HF_API_KEY"),
}

# ── Provider configs ──
ALL_PROVIDERS = [
    {
        "name": "groq",
        "base_url": "https://api.groq.com/openai/v1",
        "model": "llama-3.1-8b-instant",
        "daily_limit": 14400,
        "priority": 1,
    },
    {
        "name": "gemini",
        "base_url": "https://generativelanguage.googleapis.com/v1beta/openai/",
        "model": "gemini-2.5-flash",
        "daily_limit": 1500,
        "priority": 2,
    },
    {
        "name": "mistral",
        "base_url": "https://api.mistral.ai/v1",
        "model": "mistral-small-latest",
        "daily_limit": 1400,
        "priority": 3,
    },
    {
        "name": "cohere",
        "base_url": "https://api.cohere.com/compatibility/v1",
        "model": "command-r-plus",
        "daily_limit": 1000,
        "priority": 4,
    },
    {
        "name": "huggingface",
        "base_url": "https://api-inference.huggingface.co/v1/",
        "model": "Qwen/Qwen2.5-Coder-1.5B-Instruct",
        "daily_limit": 1000,
        "priority": 5,  # used last (slowest)
    },
]

# ── Filter to only providers where key was given ──
ACTIVE_PROVIDERS = [
    p for p in ALL_PROVIDERS
    if API_KEYS.get(p["name"]) not in (None, "")
]

# ── Usage tracker ──
usage_count = defaultdict(int)
bad_keys = set()       # permanently skip these for the session (bad key)
last_reset = time.time()

def pick_provider(exclude=None):
    global last_reset
    if exclude is None:
        exclude = set()

    if time.time() - last_reset > 86400:  # reset counts every 24h
        usage_count.clear()
        bad_keys.clear()
        last_reset = time.time()

    available = [
        p for p in ACTIVE_PROVIDERS
        if p["name"] not in exclude and p["name"] not in bad_keys
    ]
    if not available:
        return None

    return min(
        available,
        key=lambda p: (
            usage_count[p["name"]] / p["daily_limit"],  # % used
            p["priority"]                                # tiebreak by priority
        )
    )

SYSTEM_PROMPT = "You are a Minecraft coding assistant. Be brief and direct."

def respond(message, history):
    if not ACTIVE_PROVIDERS:
        yield "❌ No API keys provided! Add them in your Space's Settings → Variables and Secrets."
        return

    # Build messages once — reused across retries
    messages = [{"role": "system", "content": SYSTEM_PROMPT}]
    for exchange in history[-3:]:
        user_msg = exchange[0] if isinstance(exchange, (list, tuple)) else ""
        asst_msg = exchange[1] if isinstance(exchange, (list, tuple)) else ""
        if user_msg:
            messages.append({"role": "user", "content": user_msg})
        if asst_msg:
            messages.append({"role": "assistant", "content": asst_msg})
    messages.append({"role": "user", "content": message})

    tried = set()  # skip these for this request only

    while True:
        provider = pick_provider(exclude=tried)

        if provider is None:
            yield "❌ All providers failed or hit limits. Try again later."
            return

        print(f"Trying: {provider['name']} | uses today: {usage_count[provider['name']]}")

        client = OpenAI(
            api_key=API_KEYS[provider["name"]],
            base_url=provider["base_url"],
        )

        try:
            stream = client.chat.completions.create(
                model=provider["model"],
                messages=messages,
                max_tokens=1024,
                temperature=0.7,
                stream=True,
            )
            usage_count[provider["name"]] += 1

            response_text = ""
            for chunk in stream:
                token = chunk.choices[0].delta.content or ""
                response_text += token
                yield response_text
            return  # success — stop retrying

        except Exception as e:
            err = str(e).lower()
            tried.add(provider["name"])

            if "401" in err or "unauthorized" in err or "invalid api key" in err:
                bad_keys.add(provider["name"])  # skip for whole session
                print(f"❌ {provider['name']} bad key — skipping for session")
            elif "429" in err or "rate limit" in err or "quota" in err or "exceeded" in err:
                print(f"⚠️ {provider['name']} rate limited — trying next provider")
            else:
                print(f"⚠️ {provider['name']} error: {e} — trying next provider")

# ── Startup summary ──
print("\n=== Provider Status ===")
for p in ALL_PROVIDERS:
    key = API_KEYS.get(p["name"])
    status = "✅ Active" if key not in (None, "") else "❌ No key"
    print(f"  {p['name']:<14} {status}")
print(f"\nActive providers: {len(ACTIVE_PROVIDERS)}/{len(ALL_PROVIDERS)}")
if not ACTIVE_PROVIDERS:
    print("⚠️  WARNING: No API keys found in environment!")
print("=" * 23 + "\n")

# ── UI ──
demo = gr.ChatInterface(
    fn=respond,
    title="⛏️ Minecraft Coding Assistant",
    description=f"Multi-provider · {len(ACTIVE_PROVIDERS)} active · ~{sum(p['daily_limit'] for p in ACTIVE_PROVIDERS):,} req/day",
    examples=[
        "How do I create a custom recipe in a datapack?",
        "Write a Spigot plugin that teleports a player on command",
        "What's the syntax for a /execute command with conditions?",
    ],
    cache_examples=False,
)

if __name__ == "__main__":
    demo.launch()