Skip to main content

Playground

The Playground tab is a test conversation with a published agent. You play the customer, and the agent answers exactly as it would on a real call or run. Use it to check a change before it reaches callers, to reproduce something you saw in a session, or to hear how a new greeting sounds. See Playground in the key terms.

Tools run for real

No customer sees a playground conversation, but the agent's tools call your real systems. A refund issued in the playground is a real refund, and a booking made there is a real booking. Test tools that change data against test records, or in a Sandbox environment.

Before You Start​

  • Role: the Admin or Organisation Admin role.
  • A published agent. The Playground tab is on the agent page, which only opens once an agent has been published. To test an agent before its first publish, see Testing Before You Publish.
  • For a voice test: a microphone, and permission for the browser to use it.

Which Playground an Agent Gets​

The playground matches the kind of agent. See Frontline and Background agents.

Agent typePlaygroundHow you test
FrontlineVoiceYou call the agent from the browser and talk to it.
BackgroundTextYou send it messages, as the system that normally triggers it would.

If testing is not available for an agent, its page has no Playground tab and its actions menu has no Playground shortcut.

Testing a Draft Against the Live Version​

An agent can have a live version and a draft at the same time. The live version is what callers and integrations get. The draft holds changes that have been saved but not yet published. See Draft and live.

When the agent has a draft, a switch at the top of the Playground tab offers Live Version (with its version number) and Draft.

  • Choose Draft to test your unpublished changes.
  • Choose Live Version to hear what callers get today.

The playground opens on the live version. Switching starts a new conversation, so test the same scenario on both sides to compare them. When there is no draft, the switch is hidden and the playground always uses the live version.

Testing a draft does not publish it. Callers keep getting the live version until you select Publish. See Publishing and Versions.

Opening the Playground​

  1. Open Agents.
  2. Select the agent, then select the Playground tab.

You can also choose Playground from the agent's actions menu on the Agents page, or select Open playground on the empty Sessions over time chart in the Metrics tab.

Testing by Voice​

The voice playground is a phone call to the agent, made from your browser.

  1. Under the call settings, choose a Microphone. Browser default uses whichever microphone the browser picks.
  2. Choose how Turn taking works:
    • Live conversation: the microphone stays open and you talk naturally, as on a phone call.
    • Click to talk: select the talk button to speak and again to send, or hold Space while you talk.
  3. Select Start call. Your browser asks for microphone access the first time.
  4. Talk to the agent. The stage shows whether it is Listening, Thinking or Speaking.
  5. Select End call when you are finished. Use Mute during the call if you need a moment.

While the call runs, three cards follow it:

  • Live transcript: each line appears once someone finishes talking, not word by word. Times are counted from the start of the call.
  • Last turn: how long the agent took to reply to your last line, split into thinking and speaking time.
  • This call: whether the call is Using saved settings or Using unsaved changes, the time left, the voice, speaking speed and model, and a Reference you can quote if you need help with the call.

A test call is billed like a real call. Test calls also have a maximum length so a forgotten browser tab cannot hold a line open. A notice appears when the call has under a minute left, and a long silence ends the call early.

After the Call​

When the call ends, a summary shows the number of Turns, the Typical reply and Slowest reply times, and the number of Tools used. From there you can:

  • select Call again to start a new call
  • select Open in sessions to read the saved session in the Sessions tab
  • select Copy transcript to copy the conversation as text

The score arrives a little after the call ends. Until then the summary says there is no score yet, and the session picks it up on its own.

Voice Problems​

What you seeWhat to do
Microphone blockedAllow the microphone for this site from the browser's address bar, then select Try again.
We're not picking anything upThe microphone is connected but has been silent for 20 seconds while it was your turn. Check the input device, close anything else using it, or select Change microphone.
No microphone was foundConnect a microphone and start the call again.
The voice connection droppedStart the call again.
Voice capacity is busyToo many voice tests are running at once. End another voice test, for example in another tab, and try again.

Testing by Text​

The text playground is a chat with the agent. The badge at the top reads Evaluation mode while it waits for you.

  1. Type a message in the box at the bottom and send it, or pick one of the starter prompts, such as "What can you help me with?" or "I need to speak to a person".
  2. Read the agent's reply. Tool calls appear in the conversation as the agent makes them.
  3. Keep the conversation going for as long as the scenario needs.
  4. Select New run to end this conversation and start a fresh one.

For example, to test an ACME order-lookup agent, send the kind of message your order system would pass it, such as "Customer 4471 asks where order 98213 is", and check that it calls the order lookup tool and answers with the right status.

The panel beside the chat shows:

  • This run: the number of messages and the agent's last response time (Latency)
  • Model: the model the agent uses
  • Tools · live HTTP: the tools the agent can call, or "No tools on this agent"

If a reply fails, the badge changes to Run failed and the message offers a retry.

What a Playground Session Records​

Every playground conversation is saved as a session of type Evaluate. In the Sessions list it shows the IQ mark, a Playground badge and the word "Evaluate", so it is easy to tell apart from real traffic.

An Evaluate session records the same things as a live one: the full transcript, every tool call with its input and output, the token count, and, once the session is summarised, a score and summary. A text run is summarised when it ends, which happens when you select New run or leave the Playground tab. A voice call is summarised after you end it.

Playground sessions are included in the agent's Metrics, alongside live sessions. Filter the Sessions tab to Live when you only want to see real traffic. See Sessions.

Testing Before You Publish​

The Playground tab only exists once an agent has been published. Before that, test the agent from the builder or the editor. Both use the same test panel and test the draft.

  • In the builder: the Test It Out step for a Frontline agent, or Try It Out for a Background agent. See Building with IQ.
  • In the editor: select Test at the top of the page. The button reads Test for an agent that has never been published and Test draft for one that has, and the panel that opens beside the editor carries the same title. See The Agent Editor.

Under How to test, choose who plays the other side:

Agent typeOptionsWhat happens
FrontlineIQ calls itPick a caller scenario, or Random scenario, and IQ plays the caller while you listen.
FrontlineYou call itYou call the agent from your browser, as a customer would.
BackgroundIQ tries itIQ writes a few test scenarios. Change the message or data if you like, then select Run all.
BackgroundYou try itYou send the agent a message, as the system that normally triggers it would.

After a test, IQ reviews it and lists anything worth changing, marked Needs fixing, Worth checking or Note. Select Apply fix on a finding to apply IQ's change to the draft, then test again to check the difference. Nothing reaches callers until you publish.

What Happens Next​