Playground
The Playground is where you test an agent before it goes live. You can talk to it in your browser, chat with it, or have it call a real phone. Every tool call it makes is logged as you go, and you can edit the agent's settings next to the conversation.
Playground conversations are real conversations: they use your credits like any other call or chat, and they appear in Calls & transcripts.
Open the Playground
- From the project sidebar, choose Build → Playground, then pick an agent from the Agent list.
- Or open an agent and choose Test in Playground. The Playground opens with that agent selected.
You can test agents that are still in Draft.
Test modes
| Tab | What it does |
|---|---|
| Talk in browser | A voice call with the agent through your microphone. |
| Call a phone | The agent calls a real phone number over its own line. |
| Chat | A text conversation, streamed turn by turn. |
| Tool logs | Every tool call the agent made in this Playground session. |
All modes use the agent's prompt, tools, skills and knowledge base.
Talk in browser
- Open Talk in browser.
- Select the microphone button to start the call. Allow microphone access when your browser asks.
- Talk to the agent. The transcript appears as you speak, with tool calls inline.
While the call is live you can:
- Mute and unmute your microphone.
- Type a message to the agent in the middle of the call.
- See the call duration, the number of turns and the number of tool calls.
When the call ends, select the microphone again to start a new call with the agent's latest saved settings. If you save changes during a call, the running call keeps the settings it started with. Choose Restart voice session to test the new ones.
The call ends by itself after a stretch of silence. See Hang up after silence in Call settings.
Voice testing needs a secure (HTTPS) page and microphone permission. If your browser blocked the microphone, allow it in the site settings and try again, or use the Chat tab.
How voice tests connect
Every voice test goes through the VoisX relay, the same path your production calls and voice widget use. Your browser never connects to the voice provider directly. Unless you change the test-session settings below, a test call behaves like a real one.
Try other voices
The Test session section in the configuration panel sets the Voice model, Voice and Language for this Playground session only. The agent's saved settings don't change. The test call still goes through the VoisX relay, using the settings you picked. You can't change them while a call is live; end the call, change them, and start a new one.
Call a phone
Have the agent ring a real phone, to hear exactly what your callers will hear.
- Make sure the agent has a phone number on its Phone section. See Phone numbers.
- Open Call a phone.
- Enter the number to call in E.164 format, for example
+15550000000. - Choose Call now. The call goes out from the agent's own line.
The panel shows the call status as it moves from dialing to connected to ended, and lists your recent test calls. Choose Hang up to end the call.
Test calls are real outbound calls. They are billed as outbound minutes and follow your project's outbound calling hours. If a call can't be placed, the panel shows the reason.
Chat
- Open Chat.
- Type a message, or choose one of the Conversation starters, and send it.
- The reply streams in, with any tool calls shown inline.
Choose New conversation to start over. Conversation starters are kept for this Playground session only; add or remove them in the Test session section.
Test memory
If the agent has memory turned on, you can test as a returning customer:
- Enter a Customer id, for example
cust_8812or a phone number. - Choose the channel to act as: API, Phone, WhatsApp or Widget.
- Chat with the agent. The Memory used this turn panel shows which memories it recalled for each message and how closely each one matched.
The same customer id is used when you start a voice call in Talk in browser. A new id or channel starts a new conversation. If memory is inactive, the panel tells you why, for example because it is off for the agent or your organization hasn't allowed it.
Tool logs
The Tool logs tab lists every tool call from this session, from voice and chat. Each entry shows the tool name, its status (running, ok or error), the arguments the agent sent and the result it got back. Use it to check that the agent calls the right tool with the right inputs.
Edit the agent while you test
The configuration panel next to the conversation has the same sections as the agent page: Overview, Prompt Versions, Model & Voice, Call Settings, Tools, Skills, Knowledge Base, Channels, Escalation Policy, MCP Servers and Data Extraction. Changes you save here are saved to the agent, so you need Contributor access to make them. Choose Open in editor to go to the full agent page.
Start from a template
In the Templates section, choose Create & test on a template. A new agent is created from it in your project, and the Playground switches to it.
After testing
- Test history opens Calls & transcripts, where Playground sessions are listed with every other call, including transcripts, latency and usage.
- Get code opens the project's widget settings, where you copy the embed code for your website.