Agent Builder is where you shape how your AI agent talks to your leads — what it opens with, what it asks, who it rules out, and how it sounds. You set it up once, test it in a private sandbox, and refine it as often as you like.

You'll find it in the Customer Portal under Scripting, then Agent Builder.

https://www.loom.com/share/f06a2ed8a04b4d85b9c4bbd0a2c0dfa0

Agent Builder is in early access. Right now it's where you build your agent and test it against realistic conversation scenarios in a private playground. Nothing you do here sends messages to actual leads.

Connecting your agent to live conversations is the next step, and we'll work with you directly when your account is ready. The setup you build now is the setup that goes live — so time spent here isn't a dry run, it's the real thing.

Currently it is one agent per sub account - if you have multiple "use cases" where you ask different questions, create a new sub account for each use case. In the near future we will allow a single agent to handle multiple conversation types (use cases).

————————————————————

If you've built a Verse script before, you know the tradeoff: scripts give you precise control over every message and every branch, and that control takes real setup time.

Agent Builder starts from the other end. You tell your agent what you're trying to accomplish and give it a few rules to work within, and it handles the conversation itself — reading what someone means, deciding what to do, and writing the reply in the moment.

Neither replaces the other. A script is the right fit when you want tight control over exactly what gets said and when. Agent Builder is the right fit when you want a capable agent without building and maintaining that structure yourself.

————————————————————

Your opening hook is the first message your agent sends.

You write it yourself — it isn't generated — so your first impression is always exactly what you intended. Keep it short and specific to why you're reaching out.

The Greeting and Opt-out language is added automatically, so you don't need to include it.

Everything after the opening hook is written by your agent in the moment, based on what the person actually says back.

————————————————————

Questions are the things you want to find out during a conversation — budget, timeline, whether they're already working with someone.

Your agent doesn't run these as a checklist. It works them into the conversation naturally, keeps track of what's already been answered, and won't ask the same thing twice.

Each question gets a priority, and that priority tells your agent how hard to push.

Essential

Your agent will make sure this gets asked. Reserve it for the things you genuinely need to know before a conversation is worth your team's time.

Preferred

Your agent will try to work this in, but won't derail a good conversation to get it. This is the default for new questions, and it's the right setting for most of them.

Passive

Your agent will capture this if it comes up on its own, but won't go asking. Good for nice-to-have details.

A practical tip: mark fewer things Essential than you think you need to. Every Essential question is one more thing standing between an interested lead and a booked call. Save it for what actually changes your decision.

You can add up to 10 questions.

————————————————————

Disqualification criteria are your rules for who isn't a fit. "Only serving buyers over $400k." "We don't cover rentals." "Outside our service area."

When someone clearly meets one of these, your agent ends the conversation politely rather than continuing to work a lead you can't help.

The important part: your agent only acts on a clear, explicit match. It won't guess, and it won't disqualify someone based on a hunch or an ambiguous answer. We built it to err toward keeping a conversation going, because wrongly dropping a good lead is the more expensive mistake.

Write these as plainly as you'd explain them to a new team member on their first day. You can add up to 10.

————————————————————

Tone and voice set how your agent sounds. There are four options, and you'll see a live example of each as you choose:

WARM

"Glad you reached out — are you available to chat via text?"

POLISHED

"Thanks for the reply. Are you available to chat via text?"

CASUAL

"hey, good to hear from you — are you available to chat via text?"

UPBEAT

"Awesome! Are you available to chat via text?"

Warm is the default and works well for most businesses. Pick whichever sounds closest to how your team actually talks — the nearer it is to your real voice, the more natural the handoff feels when someone from your team picks up the phone.

Tone affects how your agent says things, not what it says or does. Switching tone won't change which questions get asked or who gets disqualified.

————————————————————

This is the piece that makes the biggest difference, and it's the one most often overlooked.

Your agent uses your published FAQs as its source of truth when someone asks a question mid-conversation. Ask about pricing, hours, your service area, or how your process works, and your agent answers using your words — then picks the conversation back up right where it left off.

Two things worth knowing.

If a question isn't covered by your FAQs, your agent won't invent an answer. It acknowledges the question and steers toward a conversation with your team, where someone can answer it properly. That's deliberate — we'd rather your agent say less than say something wrong on your behalf.

Your agent knows when you're open, and reflects that when someone asks or when they message outside your hours.

The practical takeaway: the more complete your FAQs, the more questions your agent can handle on its own, and the fewer conversations stall waiting on a person. Every FAQ you add makes your agent better at its job.

You manage FAQs separately in the Customer Portal — they aren't part of Agent Builder itself, but they're one of the biggest levers you have on how well it performs. If you only do one thing after reading this, make it filling out your FAQs.

————————————————————

The Playground sits right alongside your setup. Type a message as though you were a lead, and watch your agent respond using your current configuration. No messages go out to anyone.

Try the awkward cases, not just the easy ones. Push back on the price. Say you're already working with someone. Ask something your FAQs don't cover. Give a deliberately vague answer. Make a joke. That's where you'll learn whether your setup does what you expect.

When a conversation reaches an end point — a call booked, someone disqualified, a wrong number — you'll see a summary card showing the outcome.

Restart clears the conversation and starts fresh whenever you want.

One thing to expect: because your agent writes each reply in the moment rather than reading from a script, two runs of the same scenario won't come out identical. Try a scenario a few times before judging a change.

————————————————————

As you test, you can give feedback on any reply your agent sends.

Thumbs Up marks a reply as good — a signal of what "right" looks like for your business.

Coach lets you write a note on a specific reply, explaining what it should have said or done instead.

Add notes across as many replies as you like, then hit Submit feedback to send them all at once. Your coaching comes directly to our team along with the full conversation, so we can see the context around it.

To set expectations clearly: coaching is how you talk to us, not a switch that changes your agent. Submitting a note won't alter your setup on its own.

If you want your agent to behave differently right away, adjust your questions, disqualification criteria, tone, or FAQs — those take effect immediately. Coaching is for the things a setting can't fix, and during early access it genuinely shapes what we build next. If something feels off, tell us.

————————————————————

Each time you save, Agent Builder creates a new version.

Open History at any time to see every version, who made it, and when. Expand any one to see exactly how your agent was configured at that point. Nothing is overwritten, so you can experiment freely.

One thing to know: saving a new version restarts your test conversation, so you're always testing against your current setup. Entering edit mode and then cancelling won't disturb a conversation already in progress.

————————————————————

1. Open Scripting, then Agent Builder, in the Customer Portal.

2. Write your opening hook.

3. Add your questions and set their priorities. Be sparing with Essential.

4. Add your disqualification criteria.

5. Pick a tone.

6. Make sure your FAQs are published and cover the questions you hear most.

7. Test in the Playground — including the awkward scenarios.

8. Coach anything that doesn't land.

Access to Agent Builder is limited to Managers, Admins, and Owners.

————————————————————

Your Customer Success contact can walk you through setup, review your configuration with you, or help you think through which questions are really worth marking Essential. Reach out any time.