LAAF: running a 4 GB AI stack from a potato laptop

Our project for the Coral hackathon had two heavy parts:

• a backend built on the Coral engine (1 GB+, Rust)
• a local LLM (around 3 GB)

I couldn’t deploy them. I couldn’t even test them. My options were:

1. Own a laptop with giga specs. I don’t. 🫠
2. Verify a credit card on a cloud provider for the free tier. That failed most times.
3. Stop treating the laptop as the computer.

Remember “CNG”

• Colab (Google): free GPU, enough to run an LLM up to ~16 GB VRAM.
• Ngrok: tunnels that give a private port a public URL.
• GitHub Codespaces: a real Linux box with decent specs.

Backend: SSH into a Codespace with the GitHub CLI and run any heavy service there. Tunnel its port out through ngrok.

Model: run Ollama in Colab and expose it through a second ngrok tunnel.

Laptop: just the frontend, pointed at both URLs.

Test it all, then switch to a persistent server later.

I call it LAAF: Laptop As A Frontend. The heavy lifting goes to cloud providers, for free.