Your AI Connector Docs

Chat Playground

What Is Try Out?

Try Out is the safe testing environment where you test and refine your AI agent’s responses without affecting live conversations. Nothing you do in Try Out sends a message to a real contact or changes any active chat.

Think of it as a practice room for your AI agent. You type messages as if you were a contact, see how the agent responds, tweak its instructions right next to the chat, and try again — all without any risk. Instructions you save here become a draft: the Try Out chat uses it straight away, but real conversations keep running on your last published setup until you click Publish to live in the amber bar at the top of the editor. Rewrites produced by Optimize with AI or by a thumbs-down work the same way — test the draft here, publish it when you’re happy (see Optimize with AI below).

Try Out shows up in three places, and they’re all the same underlying test chat:

  • Inside an AI Agent, as the Try Out tab — this is where most accounts test.
  • Inside the guided onboarding wizard, as the Test & Adjust step, so you can chat with your brand-new agent before it goes live for the first time.

Accessing Try Out

From an AI Agent

  1. Open AI Agents in the sidebar, under AI Studio.
  2. Click into the Agent you want to test.
  3. Click the Try Out tab.

The test chat opens straight away and runs on the Agent itself. There’s nothing to attach and nothing to set up first — a brand-new Agent can be tested the moment you’ve written its first instruction. Each Agent keeps its own test conversation, so testing one Agent never shows up in another’s thread, and duplicating an Agent gives the copy a clean thread of its own.

Your earlier test conversation may look empty from 15 August 2026. Test conversations now sit on the Agent itself, so a thread you were part-way through before that date isn’t carried across and Try Out starts you fresh. Nothing else changes: your instructions, FAQs, knowledge and media are untouched, and optimizing your Agent from a thumbs-down you gave on an earlier reply still works exactly as before. Just send your test message again to pick up where you left off.

From a live chat (the replay loop)

The fastest way to fix a conversation that went wrong is to replay it in Try Out, adjust your agent, and see the fix apply immediately. See “Fixing a Live Chat With Try Out” below.

Fixing a Live Chat With Try Out (Replay Loop)

When a real conversation goes wrong, replay it, adjust your agent, and confirm the fix — all in one loop:

  1. Open the chat that went wrong in Chats.
  2. Click the three-dot menu at the top-right of the conversation header.
  3. Select Test In Playground. You’re taken straight to Try Out for the agent that handles that chat, with the live conversation copied into the test thread so you can replay it from exactly where it stands.
  4. Scroll to the AI turn that went wrong (or replay it as your first test message).
  5. Click the thumbs-down icon under that reply. This opens the feedback dialog you’ll use to fix it.
  6. Describe what was wrong, or edit the instructions directly in the pane beside the chat — the same pane covered below.
  7. Send the same test message again (or use the restart-from-here icon on your own message, described below) to see the agent’s new answer with the updated instructions. Adjust and re-test as many times as you need.
  8. Once the reply looks right, click Save in the instructions pane, then Publish to live in the amber bar at the top of the editor — the next real contact who asks the same question gets the fixed answer.

Note on saving: the instructions pane writes to the same place the editor’s dedicated AI Instructions step writes to, so the pane and that step can never disagree — a save here is a save there. Saves land as a draft: the Try Out chat picks the draft up immediately, and real conversations switch over only when you publish. The test conversation itself (the messages you type as the pretend contact) always stays local to Try Out and is never sent to a real contact — only the instructions you explicitly save and publish are shared.

Testing Bot Responses

Once you’re in Try Out:

  1. Type a test message in the composer, as if you were the contact asking a question.
  2. The AI generates a response based on your campaign’s current configuration.
  3. Review the response to see if it matches your expectations.

Send multiple test messages to simulate an entire conversation. This helps you verify how the agent handles:

  • Common customer questions — “What are your hours?” “How much does it cost?”
  • Booking requests — “I’d like to schedule a consultation for next week.”
  • Edge cases — “Can I get a refund?” or unusual requests your contacts might send.
  • Handoff scenarios — situations where the bot should recognize it can’t help and alert a human team member.

Tip: Don’t just test the obvious questions. Try misspellings, vague requests, and off-topic messages to see how your agent handles them. Real contacts rarely ask perfectly worded questions.

The Instructions Pane

Beside the test chat, a pane shows your agent’s AI instructions — the same fields as the editor’s AI Instructions step — so you can tweak the playbook and immediately re-test without switching pages.

  • Fields mode shows each instruction section (persona and goal, business context, conversation behavior, escalation, and so on) as its own editable box.
  • Edit as text switches to a single raw-text editor for the whole playbook, for faster bulk edits — switch back to Edit as fields any time.
  • Save only becomes active once you’ve actually changed something, and saves the change as a draft, as explained above — publish when the replies look right.
  • Hide instructions collapses the pane so the test chat takes the full width — handy once you’re just running through test messages and don’t need to edit anything.

FAQs, media, and the rest of a campaign’s configuration live on their own dedicated steps (FAQs & Knowledge, Media, and so on) rather than inside the Try Out pane itself — edit them there, then come back to Try Out and resend your test message to see the change reflected.

Media plays right inside Try Out. When the bot decides to send a Media Library item during a test conversation, the image, video, audio, or document appears in the test chat just like the contact would receive it (alongside a small “Sent the ‘…’ media” tool note). No real message leaves your account. One quirk of the sandbox: previews in old test conversations expire after 7 days and show a small “preview expired” note — ask the bot again to see the file fresh.

Making Real-Time Improvements

Try Out’s real power is its test-and-improve cycle:

  1. Send a test message and read the AI’s response.
  2. Spot an issue — maybe the tone is too formal, the answer is incomplete, or the agent doesn’t know about your new service.
  3. Edit the instructions in the pane beside the chat (or the FAQ, on the FAQs & Knowledge step).
  4. Send the same test message again and see how the response improves.
  5. Repeat until the response is exactly what you want.

Example: You ask “Do you offer free shipping?” and the agent responds “I’m not sure about shipping details.” You add a FAQ entry: “Q: Do you offer free shipping? A: Yes, we offer free shipping on orders over $50.” Send the same question again and the agent gives the correct answer.

Per-Message Actions

While the agent is idle (not currently generating a reply), each message in the test thread shows its actions beside it — no need to hover:

  • On your own (tester) messages: a pencil to edit it — this rewinds the conversation to before that message and loads its text back into the composer for you to change and resend — an arrow-path icon to restart from here, which resends the same message verbatim for a fresh AI answer, and a trash icon to delete that message together with the AI’s reply to it (so you never end up with an answer to a question that’s no longer there).
  • On the agent’s replies: a thumbs-down icon opens the feedback dialog described in the replay loop above, so you can flag a bad reply and fix the instruction that caused it. It works on any Agent you can test — there’s nothing to connect first — and the rewrite it produces lands as a draft you review and publish, the same as Optimize with AI below.

Use Start Over, in the chat panel’s header, to reset the whole test conversation and begin from a blank slate.

Starting the Conversation

If your campaign is set up to send the first message (Outgoing or Combined campaigns), Try Out’s empty state offers Have AI start, which sends the configured opening message just like it would to a real contact. If your campaign waits for the contact to speak first (Incoming), simply type your first test message as the tester. This opening-message option isn’t shown when Try Out is embedded inside an AI Agent, since opening messages are a campaign/Broadcast concept with no equivalent step in the Agent editor.

Editing the Test Contact

Click Edit Contact in the chat panel’s header to open the test contact’s details — name, email, phone, tags, and notes. Since the AI reads a contact’s fields when it replies, editing them here is how you test personalization: give the test contact a name and a tag, then ask a question that depends on it, and confirm the agent’s reply picks it up correctly.

Optimize with AI

Beyond the per-reply thumbs-down, click Optimize with AI in the chat panel’s header at any time to open the same rewrite tool without needing a specific bad reply to point at — useful when you want to improve tone or coverage generally rather than fix one exact answer.

The button is there on every Agent, with nothing to set up first: a brand-new one, a copy you just duplicated, or one that arrived in your account already built. Describe what should change and you get the same side-by-side review as before — each proposed edit shown as it reads now against what the AI wants it to say, with FAQ additions, edits and removals listed separately, and nothing applied unless you accept it.

What you accept is saved as a draft on the Agent for you to review: the Try Out chat starts using it immediately so you can test the rewrite, while your live conversations carry on unchanged. When the replies look right, click Publish to live in the amber banner at the top of the Agent’s editor — or Discard to drop the rewrite and go back to what you had. See AI Agents.

Use Cases

Fixing Incorrect Bot Responses

A contact told you the bot gave them wrong information about your pricing. Open that contact’s chat, use Test In Playground, type the same question, and see what the agent says. Then update the FAQ with the correct pricing, return to Try Out, and re-test.

Adding a New Promotion

You just launched a holiday sale and want the agent to mention it. Add the promotion to the relevant section of your instructions (for example, “We are running a 20% holiday sale through December 31st”). Test a few questions like “Are there any current deals?” and verify the agent mentions the sale naturally. Save when it looks right.

Preparing for a New Service

Your business is adding a new service next month. Before contacts start asking about it, add FAQ entries that cover common questions (pricing, availability, what’s included). Test various ways contacts might ask about it and confirm the agent answers accurately.

Testing Before You Go Live

Before switching an agent on for real contacts (or launching a broadcast), use Try Out to walk through the entire experience a contact would have — from the opening message through asking questions, booking an appointment, and handling unusual requests. This catches issues before real contacts encounter them.

Tips for Effective Testing

  • Test a variety of questions — go beyond the obvious. Try edge cases, misspellings, and unexpected requests.
  • Simulate multi-message conversations — real contacts often send several messages in a row before the bot responds. Test how the agent handles message sequences.
  • Check tone and formatting — make sure responses sound natural and match your brand voice.
  • Verify links and resources — if the agent shares links, confirm they’re correct and working.
  • Test in the contact’s language — if your campaign supports multiple languages, test questions in each one.
  • Keep notes on what you changed — if you make several adjustments in one session, jot down what changed so you can undo something specific if needed.

Need Help?

If you have questions about Try Out or need help optimizing your AI agent, reach out via our email support.