Playground
Test your agent's behavior and retrieval quality before putting it live.
The Playground, available from Dashboard → Agents → your agent → Playground, is a live chat interface for testing an agent exactly as it will behave for real visitors, without needing to embed anything.
What you can do
- Chat with the agent using the same streaming responses visitors get on the widget or hosted page.
- Browse past sessions from the sidebar, restore a previous conversation, or start a new one — playground sessions are saved locally in your browser and by conversation ID in the URL.
- See source citations under assistant replies, showing which training documents were used to generate the answer.
- Leave thumbs up/down feedback on individual replies.
- See response time for each reply.
- Inspect RAG debug info — match count, top similarity score, the retrieval threshold in use, whether hybrid search was applied, and the list of source chunks that were retrieved.
The playground isn't a separate sandbox with relaxed limits — messages sent here count against the same monthly message quota and plan limits as your live widget and hosted page.
Why this matters
Use the playground after every significant change to instructions or training data: it's the fastest way to confirm the agent is retrieving the right content and answering the way you expect, and the RAG debug panel is the best tool for diagnosing "the agent didn't use my document" issues before you troubleshoot further.