Playground

The Playground is your open canvas for prompt experimentation — a chat-like interface with mode controls, few-shot injection, and a full session history.

Overview

The Playground gives you a familiar chat interface to test prompts. Unlike a standard chatbot, it exposes every variable that affects model behavior — mode, temperature, system instructions, and few-shot examples — directly in the UI.

Every run is logged to your session history and can be rated, reloaded, or exported.

Open-Ended vs Closed-Ended mode

These two modes are the most impactful controls in the Playground. They appear as toggle buttons in the bottom toolbar.

Mode	Temperature	Best for
Open-Ended	1.2	Creative writing, brainstorming, diverse ideas
Closed-Ended	0.0	Factual Q&A, structured output, data extraction
Default	0.7	General-purpose balanced responses

Clicking a mode button sets the temperature automatically and toggles the mode on/off. Clicking an active mode resets temperature back to 0.7.

Tip

Temperature is not the only thing that changes behavior — your prompt wording matters too. "List three options" is inherently more closed-ended than "Imagine a world where…" regardless of temperature.

Few-Shot examples

Click the Few-Shot button in the bottom toolbar to open the few-shot panel. Add user/assistant pairs to show the model the exact format and style you expect.

Few-shot examples are prepended before your prompt as message history, effectively teaching the model by demonstration.

User:   What is the capital of France?
Assistant: Paris.

User:   What is the capital of Japan?
Assistant: Tokyo.

User:   What is the capital of Brazil?  ← your actual prompt

The model will follow the format established by your examples. This is especially powerful for structured extraction tasks.

System prompt

The System Instructions field in the right panel is injected as the system message at the start of every run in the current session. Use it to define the model's role, tone, and constraints.

You are a concise technical writer.
Respond in plain English, avoid jargon, and keep responses under 100 words.

Note

The system prompt persists across all messages in a session. To reset it, clear the field.

Session history

Switch to the History tab to see all runs from your account. Each entry shows the prompt, model, temperature, and output. You can:

Rate runs (1–5 stars) for future reference
Load a past run back into the editor
See token and latency metrics

Tips

Use Ctrl+Enter (or ⌘+Enter on Mac) to run without reaching for the mouse
Be specific — "summarize in 3 bullet points" gets better results than "summarize"
If the model ignores your instructions, move them to the System prompt
Try the same prompt with Open-Ended and Closed-Ended mode to understand how temperature affects creativity vs. precision

Previous← Introduction NextPrompt Tuning →