Playground

The Playground is a chat surface for testing a project's live agent exactly as an end user would experience it — the same answering pipeline, real responses — so you can try prompts, verify behaviour, and inspect what the agent did.

What you can do

• Have a back-and-forth conversation with the project's agent; start new chats or reopen previous ones.
• Watch the agent work in real time — recognising intent, searching the knowledge base, calling tools — as it produces an answer.
• Inspect each answer's details: the model used, tokens consumed, tools called, and knowledge-base passages retrieved.
• Ask for structured (JSON) answers using a schema, pass custom headers through to the tools the agent calls, and toggle streaming or extra debug detail.
• Start a voice call with the agent when voice is enabled.

Readiness

If the project isn't ready to answer (for example, no AI provider key is set up), input is blocked and you're guided to complete setup — avoiding a guaranteed failure. This is the canonical "test the agent" experience; configuring the agent's identity is done under Your Agent.