Using the Playground
Select a Provider and Model
Choose from any of your configured providers and their available models.
Features
Provider and Model Selection
Switch between any configured provider and model from a dropdown menu. This makes it easy to compare how different models respond to the same prompt.- OpenAI
- Anthropic
- Others
GPT-4o, GPT-4o-mini, GPT-4, and other OpenAI models.
Streaming Responses
Responses are streamed in real-time, token by token. You see the response as it is generated, just like in production.Session Tracking
Each conversation is saved as a session with:| Field | Description |
|---|---|
| Title | Auto-generated or custom title |
| Model | The model used for the conversation |
| Messages | Full message history with roles (system, user, assistant) |
| Token usage | Input and output token counts per message |
| Cost | Calculated cost for each message |
Conversation History
All conversations are automatically saved. Return to any previous session to:- Review past responses
- Continue the conversation
- Replay with a different model
Token and Cost Tracking
Every message in the playground shows:- Input tokens — Tokens sent in the prompt (including conversation history)
- Output tokens — Tokens generated in the response
- Cost — Calculated cost based on the model’s pricing
Use Cases
Prompt Testing
Iterate on prompt templates before deploying them to production. Test edge cases and refine system messages.
Model Comparison
Compare how different models respond to the same prompt. Evaluate quality, style, and accuracy side by side.
Debugging
Reproduce specific request patterns to debug issues. Test guardrails and policy behavior interactively.
Demos
Showcase AI capabilities to stakeholders without building a custom interface.
How It Works
The playground uses the same proxy pipeline as production API requests. This means:- Guardrails are enforced on playground requests
- Budgets are checked and deducted
- Analytics are recorded for playground usage
- Caching applies to playground requests
Playground requests are real requests that go through the full gateway pipeline. They count against your usage limits and budgets, and they appear in your analytics.
Tips
- Use system messages to set context before starting a conversation
- Compare responses by opening multiple sessions with different models
- Use the playground to test guardrails by sending content that should trigger them
- Check token counts to estimate production costs before deploying a prompt