Playground

The playground is where you find out what a model does before you write anything against it. Every run is a real call to the real gateway — there is no sandbox mode and no simulated response, so what you see is what your code will get.

Open it at console.hanzo.ai → Playground, or jump straight in from any model's row in Models.

What you can try

The model picker is the live catalogue from GET /v1/models, with each model's context window and per-million-token rate shown next to it — the same numbers the Models page renders.

What it measures

Chat runs stream, and the playground times them: time to first token and total latency, measured on the response you are watching.

Token usage comes from the gateway's own final usage chunk, so the cost shown is computed from real counts rather than an estimate. A run that fails, or that you stop halfway, keeps the partial text and reports what happened instead of throwing it away.

Take it to code

Every composer has a request preview: the exact JSON body being sent, and a runnable curl built from that same body — so the thing you copy is the thing that just ran, not a reconstruction of it.

curl https://api.hanzo.ai/v1/chat/completions \
  -H "Authorization: Bearer $HANZO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "enso-flash",
    "messages": [{ "role": "user", "content": "Name three uses for a paperclip." }],
    "temperature": 0.7
  }'

Sharing and history

The full composer state — mode, model, system prompt, messages, and settings — encodes into the URL, so sending someone a link sends them the exact setup. That is also how "try this model" links from the catalogue work.

Saved prompts and run history are stored in your browser, not on the server. They stay on the machine you created them on, and clearing site data clears them. Prompts you want to keep, version, and ship belong in Prompts.

Keys

The browser never holds your API key — the console proxies runs server-side with a short-lived token. The gateway is what enforces access: if your plan does not cover a premium model, the run is refused there with a 402 no matter what the picker offered.

Related

• Models — the catalogue the picker reads, and where to start a run
• Prompts — versioned, labelled prompts for things that ship
• AI API — the endpoints every tab calls
• Pricing — how the cost figure is derived
• API Keys — mint the key for calling from code