Playground
The Playground provides an interactive interface for testing and experimenting with AI models. Use it to explore model capabilities, test prompts, and compare responses without writing any code.
The playground supports both Generative and Embeddings models. The type of
playground that's shown for a model is determined by the model's type (shown in
the model's details page). The type is automatically determined from the name
using a heuristic - to override it, you can set the type field for a model on
the overview page.
This guide will discuss how to use the Playground effectively for generative models.
Getting Started
Select any model from your available models to begin a conversation. The Playground maintains conversation context, allowing natural back-and-forth interactions just like ChatGPT or Claude's web interface.
Each conversation is independent. Start a new conversation to reset the context or to try a different approach. You can switch between models mid-conversation to see how different models respond to the same context.
Use Cases
The Playground excels at several tasks that would be cumbersome through the API:
Prompt Development: Test and refine prompts interactively before implementing them in code. Immediate feedback helps you understand what works and what doesn't.
Model Comparison: Send identical prompts to different models to understand their strengths, weaknesses, and stylistic differences. This helps you choose the right model for production use.
Capability Exploration: Discover what models can do by experimenting with different types of requests. Try various tasks like analysis, generation, transformation, and reasoning.
Quick Tasks: For one-off tasks that don't justify writing code, the Playground provides immediate access to AI capabilities.
Conversation Management
The Playground maintains the full conversation history in the current session. This context allows models to reference earlier messages, maintain consistency, and provide relevant follow-up responses.
Long conversations consume more tokens with each request as the entire history is sent to the model. If responses become slow or you hit token limits, start a new conversation to reset the context.
Your conversations are not permanently stored. If you need to preserve important outputs, copy them before leaving the page or starting a new conversation.
Model Behavior
Different models exhibit different conversational styles and capabilities. Some models are more verbose, others more concise. Some excel at technical tasks, others at creative work. The Playground lets you experience these differences directly.
Model responses in the Playground match what you'd receive through the API with equivalent prompts. This consistency means you can prototype in the Playground and expect similar results in production.
Limitations
The Playground is designed for testing and experimentation, not production use. It has several intentional limitations:
Session-based: Conversations exist only for your current session and aren't saved.
No automation: The Playground requires manual interaction. For automated or programmatic use, switch to the API.
Tips for Effective Use
Write clear, specific prompts. Models respond better to well-defined requests than vague questions.
Use system prompts or initial instructions to set the model's behavior for the entire conversation. This is more efficient than repeating instructions in every message.
When comparing models, use identical prompts for fair comparison. Small wording changes can significantly affect responses.
If a model isn't responding as expected, try rephrasing your prompt or starting a new conversation with clearer initial instructions.
Moving to Production
Once you've developed effective prompts in the Playground, implement them in your application using the API. The model identifiers shown in the Playground are exactly what you'll use in API calls.
Remember that while the Playground is great for experimentation, production applications need proper error handling, retry logic, and potentially fallback models that the Playground doesn't provide.