Playground

What it is

The Playground is an interactive console for trying a prompt against a real (or mock) model — without writing any code. Fill in the variables, pick a provider, run, and see both the rendered prompt and the model's output.

Why it matters

It closes the loop between editing a prompt and knowing if it works. You can iterate on wording and immediately see the effect, before publishing.

Model providers

Provider	Needs
mock	Nothing — runs offline, returns a deterministic echo. Great for testing.
openai	An API key
claude	An API key
gemini	An API key
ollama	A Base URL (e.g. http://localhost:11434) for your local Ollama

API keys are entered per-run and used only for that call — they are not stored.

Using it

1. Open the Playground.
2. Optionally pick an existing prompt to load its content, or type content directly.
3. The Playground detects every and shows an input for each — fill them in.
4. Choose a model provider and model. Start with mock if you just want to see the rendered prompt with no API key.
5. Click Run.

The result panel shows:

• Rendered prompt — your content with all substituted.
• Model output — what the provider returned, with the provider and model used.

Behind the scenes

The Playground calls POST /api/playground/run. Every run is also recorded for Observability — you can see it later in the run log.

Next steps

• To save a prompt + provider + model as a reusable, runnable unit, see Agents.
• To chain multiple steps together, see Workflows.