Skip to main content
The AI Helper is a chat assistant built into the agent editor. Instead of filling out forms, you describe what you want in plain language and the AI Helper writes the configuration for you.

​
How it works

  1. You describe what you need — Tell the AI Helper what your agent should do, how it should behave, or what you want to change.
  2. It proposes changes — The AI Helper shows you exactly what it will configure as a diff card. By default nothing is saved until you approve it.
  3. You approve or refine — Accept the changes, ask for adjustments, or reject them.
Auto-apply. The console’s settings menu has a Do not confirm changes toggle. Turn it on and the AI Helper applies its edits immediately, skipping the diff-card approval step. Each applied change still appears in the chat log (marked “Auto-applied change to …”), and you can reverse any individual change from its entry there — so auto-apply speeds things up without losing the ability to undo.

​
What it can do

The AI Helper can configure most parts of your agent:
  • Write and refine the agent’s purpose and guidelines
  • Create rules
  • Generate example conversations
  • Set up pricing tiers
  • Configure hooks and memory fields
  • Write policy documents

​
What requires the dashboard UI

A few actions need to be done manually in the dashboard:
  • Renaming your agent
  • Uploading files
  • Setting up OAuth connections (connectors)
  • Connecting your Stripe account (Stripe Connect) for owner payouts
  • Installing plugins
  • Generating API keys

​
Example conversations

​
Setting up a new agent from scratch

You: I want a customer support agent for a SaaS product called Vaultly — it’s a password manager for small teams. The agent should handle billing questions, account access issues, and feature how-tos. It should be professional but friendly, and always offer to escalate to a human for anything account-security related. AI Helper: Here’s a proposed purpose and guidelines for Vaultly Support: You are the Vaultly support assistant — a friendly, professional helper for small teams using the Vaultly password manager… Proposed rules include: always offer human escalation for account security issues, never ask for master passwords, stay within Vaultly-related topics. You: Good. Add a rule that it should also escalate if a user mentions they’ve been locked out for more than 24 hours. AI Helper: Added. Updated rule: “If a user mentions being locked out of their account for more than 24 hours, always offer to escalate to the support team immediately rather than continuing to troubleshoot.”

​
Generating example conversations

You: Generate three examples showing how the agent should handle a user who is frustrated about a charge they don’t recognise. AI Helper: Here are three examples: Example 1 — charge identified as a plan upgrade: User: “I was charged 49lastweekandIhavenoideawhatit′sfor."Agent:"I′msorrytohearthat—unexpectedchargesarestressful.Letmehelpyoufigurethisout.The49 last week and I have no idea what it's for." Agent: "I'm sorry to hear that — unexpected charges are stressful. Let me help you figure this out. The 49 charge is most likely your monthly Pro plan renewal. You can see a full breakdown of your billing history in Settings → Billing → Invoices. Does that match what you’re seeing, or would you like me to walk you through it?” Example 2 — charge from a team member: User: “There’s a charge for an extra seat I didn’t authorise.” Agent: “Got it. It looks like an additional seat was added to your team — that sometimes happens when a team member accepts an invite. I can walk you through how to check who was added in Settings → Team, or if you’d like a refund for an unauthorised seat, I can connect you with our billing team directly. Which would you prefer?” You: These are good. Apply them.

​
Iterating on behaviour

You: The agent is giving answers that are too long. Make it more concise. AI Helper: I’ll add a guideline: “Keep responses short — 2-4 sentences unless the user explicitly asks for more detail. Lead with the answer, then offer to expand.” Want me to also add an example showing the desired response length?

​
Asking for a review

You: Review my current rules and tell me if anything is missing for a healthcare information agent. AI Helper: Looking at your current rules, here are gaps I’d suggest filling:
  1. No medical diagnoses — Add: “Never diagnose conditions or interpret specific symptoms as evidence of a particular disease.”
  2. Emergency referral — Add: “If a user describes symptoms that sound like a medical emergency (chest pain, difficulty breathing, signs of stroke), immediately direct them to call emergency services.”
  3. Jurisdiction disclaimer — Add: “Note that information may not apply to all countries or healthcare systems; recommend the user confirm with a local provider.”
Want me to add these?

​
Tips

  • Be specific. Instead of “make it friendly,” say “greet users by name and use casual language, but stay professional when discussing account issues.”
  • Build in stages. Start with purpose and rules, then add tools, then refine with examples.
  • Ask it to review. “Review my current rules and suggest what’s missing for a legal compliance agent.”
  • Use it for examples. Describe the conversation pattern you want: “Add an example where the user asks for a refund and the agent collects their order number before responding.”
  • Iterate freely. Unless you’ve turned on Do not confirm changes, every change is proposed before it’s applied — you can reject anything and ask for a different approach. Auto-applied changes can still be reversed from the chat log.