SMS Agent Tester
Simulate a real conversation with your SMS agent before it goes live — catch issues in the sandbox, not in front of your contacts.
Available on Growth and Scale plans.
Overview
The SMS Agent Tester is a built-in sandbox that lets you have a practice conversation with your agent using the exact same instructions and settings it would use in a live campaign. Nothing you do in the tester is sent to real contacts, and no credits are consumed.
Use it any time you create a new agent, update your system prompt, or simply want to verify that your agent is behaving the way you expect.
How It Works
Opening the tester
Navigate to your SMS agent's settings page and click Test Agent. A conversation window opens — this is your sandbox.
Simulating a conversation
Type a message in the tester as if you were a contact receiving your broadcast for the first time. For example: "Yes, I'm interested — tell me more." The agent replies using your live system prompt and configuration.
Continue the conversation naturally. Try different paths: a contact who is ready to book, a contact who has objections, a contact who asks an off-topic question. The tester lets you explore all of these without any real-world consequences.
What the tester checks
The tester runs your agent under the same conditions it would face in production:
- It uses your current system prompt exactly as written.
- It respects your max turns setting.
- It will escalate if your instructions tell it to.
If something looks wrong — the agent says something unexpected, gives a vague answer, or fails to escalate when it should — you can go back and edit your system prompt immediately without leaving the page.
Resetting the conversation
Click Reset to clear the conversation and start fresh. This is useful for testing multiple scenarios back to back without them bleeding into each other.
Tips & Best Practices
- Test before every launch. Run through at least one full conversation in the tester before attaching a new or updated agent to a live campaign.
- Simulate your most common contact types. Think about the range of people who typically respond to your campaigns and test how the agent handles each type.
- Try to confuse it. Send short, ambiguous, or off-topic replies — "maybe," "huh?", "who is this?" — to see how the agent handles unclear input. If the responses are not good enough, refine your prompt.
- Test the escalation path. Deliberately trigger escalation (for example, ask to speak with a human) to confirm the agent hands off correctly.
- Check for tone. Read the agent's replies as if you were a real contact. Does it sound natural? Friendly? On-brand? Small wording changes in your system prompt can make a big difference.
- Iterate quickly. The tester is designed for rapid iteration. Edit your prompt, reset, and test again until you are satisfied with the results.