Try it in five minutes

Three ways in, fastest first.

1. The hosted Playground (zero install)

Open the Playground, click a suggestion chip, and watch the agent build the app — then run it in the sandbox. Your user file lives in the browser (OPFS) and exports whole from settings whenever you want it.

To make built apps think, open settings and either:

• Bring your own key — paste an Anthropic or OpenAI key. It is stored in snug_secrets inside your own local user file — stripped from hub sync and default exports, sent only to the provider.
• Local model — point at any OpenAI-compatible localhost endpoint (Ollama works out of the box on desktop).

2. The desktop app (recommended)

Download Snug for macOS. Everything the Playground does — including optional passphrase encryption of your file at rest — plus what a browser structurally cannot: your file at ~/Snug on your own disk, LAN devices like Philips Hue, personal WhatsApp as a linked device, and Ollama autodetection. No account, no telemetry.

3. From source (the reference implementation)

Under ten minutes, no API key needed:

git clone https://github.com/snugprotocol/snug.git && cd snug
pnpm install                       # ~2 min
pnpm dev                           # server (mock "demo brain" — no key) + Playground, one command
                                   # open the printed URL → click a chip → build → run

Verify the whole stack if you like:

pnpm build && pnpm test            # everything green (under 3 min)
pnpm smoke                         # headless happy path via the mock adapter

The starter apps in examples/ run with zero setup and double as documentation.

Next

• Build your first app — what the agent actually produces.
• Implementor quickstart — host Snug in your own product.