Playground
The Playground is an interactive sandbox for writing and testing prompts against live models. Compare prompts, models, and parameters side by side, then save the winning configuration straight into your prompt library.
Open it from Playground in the left nav. To call models you first connect provider credentials — see Provider API Keys.
What you can do
- Compose a prompt — system + user messages with
{{variable}}placeholders. - Pick a model and parameters — provider, model, temperature, max tokens, top-p, etc.
- Run and inspect — see the output with token counts and latency.
- Compare side by side — open multiple windows to test different prompts, models, or settings at once; run them all together.
- Structured output — attach a JSON schema to constrain the model's response shape.
- Tools — include tool definitions so the model can produce tool calls.
- Save to Prompts — export a window's configuration as a new prompt (or version).
Multi-window comparison
The Playground supports multiple windows so you can run an A/B (or A/B/C…) comparison in one place — e.g. the same prompt on two models, or two prompt variants on the same model. Run them simultaneously and compare outputs, latency, and tokens directly. Window layouts persist across sessions.
A typical loop
Draft
Write your prompt and choose a model and parameters.
Compare
Duplicate the window, change one thing (model, temperature, or wording), and run both to compare.
Constrain (optional)
Add a structured-output schema or tools if your app needs them.
Save
Save the best window to the prompt library, then promote it with a label when ready.
Next steps
- Prompt Management — where saved prompts live and get versioned.
- Schemas & Tools — define structured outputs and tools.
- Provider API Keys — connect the models the Playground calls.