Using the Prompt Playground
Test prompts against different models with adjustable parameters before deploying them as routing rules.
Prerequisites
- A Xilos account with access to the dashboard
- At least one LLM configured in Model Config
What Is the Prompt Playground?
The Prompt Playground (formerly Rules Sandbox) lets you test how different models respond to specific prompts with different parameters — without creating a routing rule or affecting production traffic. It's a safe space to experiment with:
- Different LLM models
- Temperature settings
- Max token limits
- System prompts
- Query variations
Step 1: Access the Playground
- Navigate to Prompt Playground in the sidebar.
- You'll see a chat interface with parameter controls.
Step 2: Set Parameters
Configure the test parameters:
- Model — Select any available model from the dropdown
- Temperature — Adjust the slider (0 = deterministic, 2 = creative)
- Max Tokens — Set the maximum response length
- System Prompt — Write or paste a system prompt to test
Step 3: Enter a Test Query
Type a query in the input field and submit. The response shows:
- The model's response
- Token count (input + output)
- Estimated cost
- Latency
Step 4: Compare Models
- Send the same query with different models selected.
- Compare response quality, cost, and latency.
- Use this to decide which model to assign in your routing rule.
Step 5: Test System Prompts
- Write different system prompt variants.
- Send the same query with each variant.
- Identify which prompt produces the best results.
- Use the winning prompt when creating your routing rule.
Info: The Prompt Playground does not consume your production query log — test queries are isolated and don't affect dashboards or cost tracking.
Best Practices
- Test with realistic queries that represent your actual use case
- Try multiple temperature settings — some tasks benefit from creativity, others need determinism
- Test edge cases — ambiguous queries, short queries, long queries
- Compare at least 2-3 models before deciding
- Save the best system prompt for use in your routing rule
- Use the results to set appropriate max_tokens limits