Playground

The Playground sends real data-plane requests through DeltaLLM using an API key you provide. Use it to validate a configured model, routing, key scope, and request parameters before integrating an application.

Access and prerequisites

• Any authenticated Admin UI account can open Playground.
• Execution requires a DeltaLLM API key; the browser session is not substituted for gateway auth.
• The key must be allowed to call the selected public model/route target and remain within its budget and rate limits.
• At least one deployment with the relevant workload mode must exist. The selector groups chat, text-to-speech (audio_speech), and speech-to-text (audio_transcription) deployments.

Do not paste a production application's credential into an untrusted browser or shared screen. Create a short-lived, narrowly scoped test key when possible.

The model status can be Online, Offline, or Unknown. Unknown means that health data is unavailable. It does not mean that the model is offline. If the model list fails to load, select Retry.

Chat

1. Select Chat, a healthy chat/completion deployment, and enter the test API key.
2. Open Parameters to set the system prompt, temperature, maximum tokens, top-p, and penalties.
3. Send a message. The response streams through /v1/chat/completions.
4. Open Stats for token counts and latency; open Inspect for the request shape and a curl example.
5. Use Stop to abort generation or Reset to clear the local conversation.

Success means a streamed response finishes, Stats identifies the selected public model, and the request appears under Usage & Spend for the key's authorized scope.

Text to speech

1. Select Text to Speech and an audio_speech deployment.
2. Enter text, choose an available voice, speed, and output format, then generate.
3. Play or download the returned audio.

Success means the browser receives playable audio from /v1/audio/speech and reports request latency/file size. An empty model selector means no compatible deployment is configured. If the browser cannot play the returned audio, the Playground shows the browser's reason; use Download to keep the file.

Speech to text

1. Select Speech to Text and an audio_transcription deployment.
2. Upload an audio file up to 25 MB or record from the browser after granting microphone access.
3. Optionally set language, response format, and prompt; then transcribe.

Success means /v1/audio/transcriptions returns text and the Playground shows word count and latency. Browser recording requires a secure context in normal production browsers.

Troubleshooting

Symptom	Check
Page visible but Send/Generate is disabled	Enter an API key and select a compatible deployment
No models for one mode	Create a deployment with the matching model_info.mode
401 or 403	Key validity, expiry, owner scope, and callable-target access
429	Key/team/org/tier limits or budget and the response rate-limit headers
Provider error	Deployment health, provider credential, upstream model ID, and fallback events
Microphone unavailable	Browser permission, secure origin, device policy, and supported recording format

See Models, API keys, and the proxy endpoint reference.