Turn any old laptop into an AI server with a dedicated free API key.

Oldy is a lightweight wrapper for Ollama designed specifically for recycling old hardware into accessible AI endpoints.

pip install oldy

oldy start      # Auto-detect hardware, pick model, and start public server
oldy stop       # Shut down the tunnel and local engine
oldy status     # Check RAM, CPU, and model performance
oldy models     # Browse and switch between supported models
oldy logs       # View live request logs
oldy url        # Display your public ngrok URL
oldy key        # Display your secure API key

While Ollama is the engine, Oldy is the specialized driver for aging hardware:

• Hardware Safety: Ollama will let you download models that crash your system; Oldy calculates your RAM/CPU first to ensure a stable experience.
• Instant Public Access: Ollama only works on localhost; Oldy automatically provisions a secure public URL and API key.
• Hardware Monitoring: Built-in status checks help you see exactly how the model is performing on your specific hardware.

Oldy is compatible with the OpenAI API format. Just provide your public URL as the base_url and your generated API key.

from openai import OpenAI

client = OpenAI(
    base_url="YOUR_OLDY_URL/v1",
    api_key="YOUR_API_KEY"
)

response = client.chat.completions.create(
    model="gemma:2b",
    messages=[{"role": "user", "content": "Hello!"}]
)
print(response.choices[0].message.content)

curl -X POST YOUR_OLDY_URL/api/generate \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "gemma:2b", "prompt": "Hello!", "stream": false}'

Oldy manages Ollama under the hood so you don't have to. An ngrok tunnel makes your local server publicly accessible without networking setup. All traffic is secured by a built-in FastAPI proxy that handles:

• Authentication: Only requests with your Bearer token are allowed.
• Rate Limiting: Enforces a 30 req/min limit to prevent hardware overheating.
• Header Sanitization: Ensures compatibility between public requests and the local engine.

more models coming soon

Requirements: Python 3.10+, Linux/macOS/Windows, 4GB+ RAM.