Runtype
Playground

What is the Playground?

The Playground is your space to experiment with prompts, models, and settings with no setup required. It is the fastest way to find the right combination before committing to a Flow or Agent.

Think of it as a scratchpad where you can iterate freely, test edge cases, and build confidence in your AI configuration before it goes live.

When to use the Playground

  • Refine prompts — Try different wording, instructions, and formats to see what gets the best results.

  • Compare models — Run the same prompt through different models to find the best fit for your use case. If you want to compare model behavior more systematically, see What are Evals?

  • Tune settings — Adjust temperature, max tokens, and other parameters to dial in the behavior you need.

  • Test tools — Attach Tools and see how a model uses them before building a full workflow.

  • Share quick demos — Show stakeholders real AI responses before investing time in a full Flow or Agent.

Example

You are building a customer support chatbot that handles refund requests. Before adding the prompt to a Flow, test it in the Playground:

  1. Start with gpt-5 and a system prompt that covers your refund policy.

  2. Test edge cases such as angry customers, missing order numbers, and requests outside the return window.

  3. Switch to claude-sonnet-4-6 and compare how each model handles the same scenarios.

  4. Lower the temperature to 0.3 for more consistent, policy-aligned replies.

  5. Once you are happy with the results, click Save as Flow to turn it into a production workflow.

This kind of rapid iteration helps you catch weak prompts and edge-case issues early, before they reach your users.

Opening the Playground

Click Playground in the left sidebar. The interface opens immediately with your default model selected, so you can start testing right away.

Playground interface

The Playground has three main areas.

Model selection

Choose the model you want to test from the dropdown. The models available to you depend on whether you are using platform keys or your own API keys (BYOK).

Not sure which model to start with? Try a general-purpose model like gpt-5 or claude-sonnet-4-6 first, then experiment from there.

Prompt editor

Write your system prompt in the editor. You can include template variables like {{customer_name}} or {{order_id}}. The Playground will ask you to fill in values before running.

Template variables are a good way to test how your prompt handles different inputs without rewriting it each time.

Advanced Settings

Click Advanced Settings to fine-tune model behavior:

  • Response Format — Choose between the model's default output or structured JSON.

  • Temperature — Controls randomness. 0 is focused and deterministic. 2 is highly creative.

  • Max Tokens — Set a response length limit, or leave it at the model default.

  • Reasoning — Enable or disable the model's reasoning behavior on supported models.

Start with a temperature around 0.7 for most use cases. Go lower, around 0.2 to 0.3, when you need consistent responses such as customer support or data extraction. Go higher, around 0.8 to 1.0, for creative tasks such as brainstorming or content generation.

Running a test

  1. Select a model.

  2. Write your system prompt and add any template variables you need.

  3. Optionally adjust Advanced Settings.

  4. Optionally attach Tools to test tool-calling behavior.

  5. Type a message in the chat panel and send it.

  6. Review the response, timing, and token usage shown below the reply.

You can keep chatting to test multi-turn conversations. Each message builds on the previous context, just like it would in a deployed Flow or Agent.

Saving your configuration

Once you have found a prompt and model combination that works, you can save it directly from the Playground without rebuilding anything from scratch.

  1. Click Save as Flow or Save as Agent.

  2. Name your Flow or Agent and confirm.

Use Save as Flow when you have a well-tuned prompt that is ready to become a step in a workflow. Your model, prompt, and settings carry over. If you want to keep building, see Creating and editing Flows.

Use Save as Agent when you need multi-step reasoning with tool use. Your tool attachments and configuration are preserved so the Agent can pick up where you left off. For the next step, see Creating and configuring Agents.

Next steps

Was this helpful?