Playground is a temporary test environment for the selected agent. Prompt experiments do not change the live bot, and Playground conversations are not saved.
Run a safe test
- Select an agent and open Playground.
- Choose a model and reasoning effort, then adjust the system prompt, guardrails, fallback response, and conversation starters.
- Turn Knowledge on or off depending on what you want to isolate.
- Send test messages. Setting changes take effect on the next message.
- Select Reset settings to restore the agent’s saved configuration.
What stays temporary
- System prompt, model, reasoning effort, guardrails, fallback text, and starter changes are not saved.
- Playground messages do not appear in the live Conversations inbox.
- The public widget’s saved appearance and behavior are not changed.
Actions are the exception
If the agent has actions and your plan allows them, you can enable Actions for a test. Chatwith warns you because the conversation is temporary but an external action is real—it may create, update, send, or delete data.
Leave Actions off for prompt-only tests. If you enable them, use a sandbox account or non-production data.
Usage
Playground requests use the selected model and consume message credits. Check Account → Usage when comparing higher-cost models.