Guardrails

Hard rules your agent can’t break

Safety shouldn’t be an enterprise SKU. Every Flowyte agent ships with five deterministic guardrails — enforced on every turn, on voice and chat — which means the agent literally cannot cross the lines you set. Here they are, in full.

Free credits included. No credit card required.

Scripted demo — a replay of the real product, not a live session.

The five policies

Deterministic rules, not vibes

Most platforms claim safety in one sentence and gate the actual controls behind an enterprise contract. This is our entire Guardrails screen, restyled for this page. Values below are illustrative — you set your own.

5 guardrails active

Verify before disclose

Which fields the agent may read back, and to whom

Verify before disclose is enabled
Business hours & service areaAnyone
Appointment date & timeVerified callers only
Invoice & balance detailsVerified callers only

Caller verification methods are configured on the same screen — you choose what counts as verified.

Which means the agent can read an appointment back to the account holder — and only the account holder. Everyone else hears exactly what you decided everyone can hear.

Writes gate

Nothing changes in your systems without a gate

Writes gate is enabled
  • Confirm before any write
  • Require a verified caller

Always escalate to a human

issue_refundclose_account

Which means the agent can book, update, and log — but it confirms first, and the actions you flag always land with a human instead.

Rate limits

Ceilings on how fast and how much it can act

Rate limits is enabled
Tool calls per minute6
Tool calls per call20

Which means even a strange call can only do so much — the agent has a hard budget for actions, per minute and per call.

Business hours

A weekly schedule, plus after-hours behavior

Business hours is enabled
Mon–Fri8:00 AM–6:00 PM
Saturday9:00 AM–2:00 PM
SundayClosed

After hours: explain the schedule, take a message, and offer the first opening tomorrow.

Which means the agent answers 24/7, but behaves the way you want at 2 AM — it knows the difference between open and closed.

Content & escalation

Banned topics, PII redaction, and the human threshold

Content & escalation is enabled

Banned topics

Competitor pricingLegal adviceRefund exceptions
Redact PII from transcripts & logs
Offer a human after frustrated turns2

The escalation threshold: how many negative-sentiment turns before the agent offers a person.

Which means the topics you ban never come up, transcripts stay clean of personal data, and a frustrated caller reaches a person before they reach their limit.

When it doesn’t know

It says so — and hands the call to a human

No AI answers everything, and we won’t pretend ours does. When a caller asks something outside the agent’s knowledge or its rules, it doesn’t improvise. It says it can’t help with that, and warm-transfers to a person with the full context of the conversation — so the caller never repeats themselves. The question it couldn’t answer shows up the next morning in Observe, ranked by how often it comes up, ready to fix. For the edge cases, that handoff builds more trust than any claim of omniscience would.

The last gate before live

Read what it will say — and won’t — before anyone hears it

Guardrails aren’t a settings page you configure once and hope about. Every publish opens a pre-publish report: what the agent will say, what it won’t, exactly which data leaves your system, and how much of your knowledge it covers. Watch the real publish flow below.

ACT 4 — PUBLISH: Review what it will and won’t say, then go live with one click.

Scripted demo — a replay of the real product, not a live session.

FAQ

Guardrails, answered plainly

What are guardrails in Flowyte?
Guardrails are deterministic rules the agent cannot break — not prompts or suggestions. They are enforced on every turn of every conversation, on both voice and chat: what the agent may disclose and to whom, when it must confirm before acting, how fast it can call tools, when it answers at all, and which topics it never touches.
Are guardrails included on every plan?
Yes. All five guardrail policies — verify-before-disclose, the writes gate, rate limits, business hours, and content and escalation rules — ship with every Flowyte agent, including pay-as-you-go. Safety is not an enterprise add-on.
What happens when the agent doesn’t know the answer?
It says so and hands off. Instead of improvising, the agent warm-transfers to a human with the full context of the conversation, so the caller never repeats themselves. The unanswered question then shows up in Observe, ranked by how often it comes up, so you can fix the gap.
Can the agent leak customer data?
The verify-before-disclose matrix controls exactly which fields the agent may read back and which require a verified caller first. You can also redact PII from transcripts and logs, and before every publish the pre-publish report shows exactly what data leaves your system — which Skill sends which fields where.
How do I know what the agent will say before it goes live?
Every publish starts with a pre-publish report: a Will-say and Won’t-say review side by side, the data that leaves your system, and your knowledge coverage. Nothing goes live until you have read it and clicked publish.

Set the lines. Then let it answer.

Draft an agent in minutes, read its guardrails, and publish when you’re ready.

Free credits included. No credit card required.