Skip to main content
The playground is an interactive chat interface in the dashboard for trying models, tuning parameters, and prototyping prompts, without writing code.
The playground runs in the dashboard. Every run is metered and billed exactly like a production API call, so what you see in the playground matches what your integration will cost.

What you can do

Try any model

Pick any model in your catalog and chat with it.

Tune parameters

Adjust temperature, max tokens, and other generation settings live.

Tools & JSON

Test tool/function calling and structured (JSON) output.

See real cost

Each response shows its tokens and exact cost, deducted from your wallet.

Logged traffic, without writing code

When request logging is on for your workspace, playground chats are logged exactly like production traffic: the exchange goes through the gateway’s own pipeline: secret-scrubbed, stamped with your retention window, and visible in Logs within seconds. Each response shows a quiet “Logged to your workspace” chip once the log has actually landed (and an honest note instead when logging is off or the log failed; a logging hiccup never affects the chat itself). Set a Tag on your playground session (“support”, “extraction”, …) and the traffic is labeled everywhere tags work: filters, datasets, judge populations. Different system prompts form separate segments automatically. This makes the playground the fastest way to try the whole improvement workflow before integrating: chat a few times → grade the exchanges in Review → create and calibrate a judge → run a comparison, all on traffic you generated in the dashboard.

From playground to code

Once a prompt works in the playground, reproduce it in your app with the same model id and parameters; the playground uses the same inference path as the API. See Chat completions to move it into code.

Billing

Playground usage draws from your prepaid wallet just like API usage. It’s the easiest way to sanity-check a model’s cost and behavior before wiring it into production.