/dashboard/playground) is an in-dashboard chat UI that sends real requests through your activated gateway. Use it to smoke-test a provider after connecting it, debug guardrails, or compare models side-by-side.
What you can configure
| Control | What it does |
|---|---|
| Provider | Filters the model list to one provider. |
| Model | Any enabled chat model from the selected provider and gateway. |
| System prompt | Optional textarea. Sets the system message. |
| Temperature | 0–2. Default 1. Lower is more deterministic. |
| Max tokens | Upper bound on completion length. |
| Top-P | Nucleus sampling. |
| Streaming | Toggle. When on, tokens arrive over Server-Sent Events. |
Requirements
- At least one Online gateway (see Activate)
- At least one attached provider with an API key (see Connect a provider)
- At least one enabled model (see Sync models)
Tips
Test guardrails
Test guardrails
Turn on PII detection under Configuration → Guardrails, then paste a prompt containing a fake SSN into the playground. The gateway should block or redact it depending on policy — and the block will show up in Logs & Traces as a guardrail step.
Compare models
Compare models
Open the playground in two tabs, pick different models, send the same prompt. The trace view lets you compare latency and cost after the fact.
Copy as curl
Copy as curl
The Request JSON tab gives you everything you need to reproduce the call from
curl or your own SDK against the gateway’s /v1/chat/completions endpoint.Next step
Inspect logs and traces
Every playground request is logged and traced. See how to read them.