grelay/app.py

from flask import Flask, request, jsonify, render_template_string
from datetime import datetime
import json
import os
import requests
import traceback

app = Flask(__name__)

# Store recent requests in memory
recent_requests = []
MAX_REQUESTS = 50

# Configuration with better defaults
OLLAMA_ENDPOINT = os.getenv("OLLAMA_ENDPOINT", "http://localhost:11434")
OLLAMA_MODEL = os.getenv("OLLAMA_MODEL", "llama2")  # More common default
VIDEO_FORMAT = os.getenv("VIDEO_FORMAT", "skip")  # Skip videos by default for debugging


def test_ollama_connection():
    """Test connection to Ollama and get available models"""
    try:
        # Try to get model list
        response = requests.get(f"{OLLAMA_ENDPOINT}/api/tags", timeout=5)
        if response.status_code == 200:
            models = response.json().get("models", [])
            model_names = [m.get("name", "") for m in models]
            print(f"✅ Connected to Ollama at {OLLAMA_ENDPOINT}")
            print(f"Available models: {model_names}")
            return True, model_names
        else:
            print(f"❌ Ollama responded with {response.status_code}: {response.text}")
            return False, []
    except Exception as e:
        print(f"❌ Cannot connect to Ollama at {OLLAMA_ENDPOINT}: {e}")
        return False, []


def convert_gemini_to_ollama_generate(gemini_request):
    """Convert Gemini format to Ollama /api/generate format"""
    try:
        contents = gemini_request.get("contents", [])

        # Extract text parts and combine
        prompt_parts = []
        images = []

        for content in contents:
            parts = content.get("parts", [])

            for part in parts:
                if "text" in part:
                    prompt_parts.append(part["text"])
                elif "inline_data" in part:
                    inline = part["inline_data"]
                    mime_type = inline.get("mime_type", "")
                    data = inline.get("data", "")

                    if VIDEO_FORMAT == "skip" and (
                        mime_type.startswith("video/") or mime_type.startswith("image/")
                    ):
                        prompt_parts.append(
                            f"[Media content ({mime_type}) was present but skipped for debugging]"
                        )
                        print(f"⏭️ Skipping media: {mime_type}")
                    elif mime_type.startswith("image/"):
                        images.append(data)
                        print(f"🖼️ Adding image: {mime_type}")

        # Build Ollama request
        ollama_request = {
            "model": OLLAMA_MODEL,
            "prompt": " ".join(prompt_parts) if prompt_parts else "Hello",
            "stream": False,
            "options": {},
        }

        # Add images if present and not skipping
        if images and VIDEO_FORMAT != "skip":
            ollama_request["images"] = images

        # Add generation config
        gen_config = gemini_request.get("generationConfig", {})
        if "temperature" in gen_config:
            ollama_request["options"]["temperature"] = gen_config["temperature"]
        if "maxOutputTokens" in gen_config:
            ollama_request["options"]["num_predict"] = gen_config["maxOutputTokens"]

        return ollama_request
    except Exception as e:
        print(f"❌ Error converting to Ollama format: {e}")
        raise


def convert_ollama_generate_to_gemini(ollama_response):
    """Convert Ollama /api/generate response to Gemini format"""
    try:
        print(
            f"🔄 Converting Ollama response: {json.dumps(ollama_response, indent=2)[:500]}..."
        )

        # Ollama /api/generate returns: {"response": "text", "done": true, ...}
        response_text = ollama_response.get("response", "")

        if not response_text:
            print(f"❌ No 'response' field in Ollama response")
            print(f"Full response keys: {list(ollama_response.keys())}")
            return {"error": "No response text from Ollama"}

        print(f"✅ Found response text: {response_text[:100]}...")

        # Convert to Gemini format
        gemini_response = {
            "candidates": [
                {
                    "content": {"parts": [{"text": response_text}], "role": "model"},
                    "finishReason": "STOP",
                    "index": 0,
                }
            ],
            "usageMetadata": {
                "promptTokenCount": ollama_response.get("prompt_eval_count", 0),
                "candidatesTokenCount": ollama_response.get("eval_count", 0),
                "totalTokenCount": ollama_response.get("prompt_eval_count", 0)
                + ollama_response.get("eval_count", 0),
            },
        }

        return gemini_response
    except Exception as e:
        print(f"❌ Error converting Ollama response: {e}")
        print(f"Ollama response was: {ollama_response}")
        traceback.print_exc()
        return {"error": f"Conversion error: {str(e)}"}


HTML_TEMPLATE = """
<!DOCTYPE html>
<html>
<head>
    <title>Ollama Proxy Debug</title>
    <style>
        body { font-family: Arial, sans-serif; margin: 20px; background: #f5f5f5; }
        h1 { color: #333; }
        .config { background: #e3f2fd; padding: 15px; margin: 10px 0; border-radius: 5px; }
        .status { padding: 10px; margin: 10px 0; border-radius: 5px; }
        .status.connected { background: #c8e6c9; }
        .status.error { background: #ffcdd2; }
        .request { 
            background: white; 
            padding: 15px; 
            margin: 10px 0; 
            border-radius: 5px;
            box-shadow: 0 2px 4px rgba(0,0,0,0.1);
        }
        .timestamp { color: #666; font-size: 0.9em; }
        .method { 
            display: inline-block;
            padding: 3px 8px;
            background: #4CAF50;
            color: white;
            border-radius: 3px;
            font-weight: bold;
        }
        .forwarded { 
            display: inline-block;
            padding: 3px 8px;
            background: #2196F3;
            color: white;
            border-radius: 3px;
            font-size: 0.8em;
            margin-left: 10px;
        }
        .error-badge { 
            display: inline-block;
            padding: 3px 8px;
            background: #f44336;
            color: white;
            border-radius: 3px;
            font-size: 0.8em;
            margin-left: 10px;
        }
        pre { 
            background: #f4f4f4; 
            padding: 10px; 
            border-radius: 3px;
            overflow-x: auto;
            max-height: 300px;
            overflow-y: auto;
        }
        .clear-btn {
            background: #f44336;
            color: white;
            border: none;
            padding: 10px 20px;
            border-radius: 5px;
            cursor: pointer;
            margin: 10px 0;
        }
        .clear-btn:hover { background: #d32f2f; }
    </style>
    <script>
        function clearRequests() {
            fetch('/clear', { method: 'POST' })
                .then(() => location.reload());
        }
        // Auto-refresh every 5 seconds
        setTimeout(() => location.reload(), 5000);
    </script>
</head>
<body>
    <h1>🦙 Ollama Proxy Debug Interface</h1>
    
    <div class="config">
        <strong>Configuration:</strong><br>
        Ollama Endpoint: <strong>{{ endpoint }}</strong><br>
        Model: <strong>{{ model }}</strong><br>
        Video Format: <strong>{{ video_format }}</strong>
    </div>
    
    <div class="status {{ status_class }}">
        <strong>Ollama Status:</strong> {{ status_message }}<br>
        {% if models %}
        <strong>Available Models:</strong> {{ models|join(', ') }}
        {% endif %}
    </div>
    
    <p><strong>Send requests to:</strong> http://localhost:5000/webhook</p>
    <button class="clear-btn" onclick="clearRequests()">Clear All</button>
    
    <div id="requests">
        {% for req in requests %}
        <div class="request">
            <div>
                <span class="method">{{ req.method }}</span>
                <span class="timestamp">{{ req.timestamp }}</span>
                {% if req.forwarded %}
                <span class="forwarded">SENT TO OLLAMA</span>
                {% endif %}
                {% if req.error %}
                <span class="error-badge">ERROR</span>
                {% endif %}
            </div>
            <div><strong>Path:</strong> {{ req.path }}</div>
            {% if req.body %}
            <div><strong>Incoming (Gemini Format):</strong></div>
            <pre>{{ req.body }}</pre>
            {% endif %}
            {% if req.ollama_request %}
            <div><strong>Sent to Ollama:</strong></div>
            <pre>{{ req.ollama_request }}</pre>
            {% endif %}
            {% if req.ollama_response %}
            <div><strong>Ollama Raw Response:</strong></div>
            <pre>{{ req.ollama_response }}</pre>
            {% endif %}
            {% if req.response %}
            <div><strong>Final Response (Gemini Format):</strong></div>
            <pre>{{ req.response }}</pre>
            {% endif %}
            {% if req.error %}
            <div><strong>Error Details:</strong></div>
            <pre style="color: red;">{{ req.error }}</pre>
            {% endif %}
        </div>
        {% endfor %}
    </div>
</body>
</html>
"""


@app.route("/")
def index():
    connected, models = test_ollama_connection()
    status_class = "connected" if connected else "error"
    status_message = "Connected ✅" if connected else "Cannot connect ❌"

    return render_template_string(
        HTML_TEMPLATE,
        requests=reversed(recent_requests),
        endpoint=OLLAMA_ENDPOINT,
        model=OLLAMA_MODEL,
        video_format=VIDEO_FORMAT,
        status_class=status_class,
        status_message=status_message,
        models=models,
    )


@app.route("/webhook", methods=["POST", "PUT", "PATCH"], defaults={"subpath": ""})
@app.route("/webhook/<path:subpath>", methods=["POST", "PUT", "PATCH"])
def webhook(subpath):
    """Accept requests and forward to Ollama /api/generate"""
    timestamp = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
    full_path = request.full_path if request.query_string else request.path

    print(f"\n{'='*60}")
    print(f"[{timestamp}] INCOMING {request.method} {full_path}")
    print(f"{'='*60}")

    # Get request data
    try:
        gemini_request = request.get_json() if request.is_json else {}
        body_display = (
            json.dumps(gemini_request, indent=2) if gemini_request else "No JSON body"
        )
    except Exception as e:
        body_display = f"Error parsing JSON: {e}"
        gemini_request = {}

    print(f"Request body: {body_display[:300]}...")

    # Store request info
    req_info = {
        "timestamp": timestamp,
        "method": request.method,
        "path": full_path,
        "body": body_display,
        "forwarded": False,
        "response": None,
        "error": None,
        "ollama_request": None,
        "ollama_response": None,
    }

    try:
        if gemini_request:
            print(f"\n🔄 Converting to Ollama format...")

            # Convert to Ollama format
            ollama_request = convert_gemini_to_ollama_generate(gemini_request)
            req_info["ollama_request"] = json.dumps(ollama_request, indent=2)

            print(f"Ollama request: {json.dumps(ollama_request, indent=2)}")

            # Send to Ollama
            print(f"\n📤 Sending to Ollama: {OLLAMA_ENDPOINT}/api/generate")

            response = requests.post(
                f"{OLLAMA_ENDPOINT}/api/generate",
                json=ollama_request,
                headers={"Content-Type": "application/json"},
                timeout=120,
            )

            print(f"📥 Ollama response status: {response.status_code}")

            if response.status_code == 200:
                ollama_response = response.json()
                req_info["ollama_response"] = json.dumps(ollama_response, indent=2)

                print(f"✅ Ollama responded successfully")

                # Convert back to Gemini format
                gemini_response = convert_ollama_generate_to_gemini(ollama_response)

                if "error" in gemini_response:
                    req_info["error"] = gemini_response["error"]
                    req_info["response"] = json.dumps(gemini_response, indent=2)
                else:
                    req_info["forwarded"] = True
                    req_info["response"] = json.dumps(gemini_response, indent=2)
                    print(f"✅ Conversion successful")

                recent_requests.append(req_info)
                if len(recent_requests) > MAX_REQUESTS:
                    recent_requests.pop(0)

                return jsonify(gemini_response), 200

            else:
                error_text = response.text
                error_msg = f"Ollama returned {response.status_code}: {error_text}"
                print(f"❌ {error_msg}")

                req_info["error"] = error_msg
                req_info["ollama_response"] = error_text
                req_info["forwarded"] = True

                recent_requests.append(req_info)
                if len(recent_requests) > MAX_REQUESTS:
                    recent_requests.pop(0)

                return (
                    jsonify(
                        {
                            "error": {
                                "message": error_text,
                                "status": response.status_code,
                            }
                        }
                    ),
                    response.status_code,
                )
        else:
            req_info["error"] = "No JSON body received"
            recent_requests.append(req_info)
            if len(recent_requests) > MAX_REQUESTS:
                recent_requests.pop(0)

            return (
                jsonify({"status": "error", "message": "No JSON body to process"}),
                400,
            )

    except Exception as e:
        error_msg = f"Exception: {str(e)}"
        print(f"❌ {error_msg}")
        traceback.print_exc()

        req_info["error"] = error_msg
        recent_requests.append(req_info)
        if len(recent_requests) > MAX_REQUESTS:
            recent_requests.pop(0)

        return jsonify({"error": {"message": error_msg}}), 500


@app.route("/clear", methods=["POST"])
def clear():
    """Clear all stored requests"""
    recent_requests.clear()
    return jsonify({"status": "cleared"}), 200


@app.route("/test")
def test_endpoint():
    """Test endpoint to verify Ollama connection"""
    connected, models = test_ollama_connection()
    return jsonify(
        {"connected": connected, "endpoint": OLLAMA_ENDPOINT, "models": models}
    )


if __name__ == "__main__":
    print("🦙 Ollama Proxy Server Starting...")
    print(f"🔗 Ollama Endpoint: {OLLAMA_ENDPOINT}")
    print(f"🤖 Model: {OLLAMA_MODEL}")
    print(f"📹 Video Format: {VIDEO_FORMAT}")
    print("📍 Web UI: http://localhost:5000")
    print("📮 Webhook: http://localhost:5000/webhook")
    print("🧪 Test: http://localhost:5000/test")

    # Test connection on startup
    connected, models = test_ollama_connection()
    if not connected:
        print("\n⚠️  WARNING: Cannot connect to Ollama!")
        print("Please check:")
        print("1. Is Ollama running? (ollama serve)")
        print("2. Is it on the right port?")
        print("3. Set OLLAMA_ENDPOINT env var if different")

    app.run(host="0.0.0.0", port=5000, debug=True)