How It Works

Inline guidance, not another tool

Guardrail CX runs quietly inside the browser. It reads what's already on screen, highlights what matters, and reviews drafts before they're sent. No integrations, no new workflows, no data leaving the page.

Guardrail CX providing inline guidance on a draft

Browser-Native

Runs entirely in the browser. Reads visible text on the page — no API calls, no backend, no integrations required.

Zero Configuration

Auto-detects CX platforms by analyzing page structure. Agents don't need to set anything up or change their workflow.

Privacy by Design

All analysis happens locally. No customer data is transmitted, stored, or logged outside the browser tab.

Step by Step

Four layers of quiet guidance

Each layer works together to catch risks early, surface them gently, and give managers the controls they need — without disrupting the agent's workflow.

Reads the active conversation

When an agent opens a conversation, Guardrail CX automatically identifies the active thread by locating the draft input area. It reads the visible customer messages — the last 3 to 8 messages in the thread — and analyzes them for escalation signals.

Automatically detects the active thread from the focused draft area

Reads only visible, on-screen text — no API calls, no data extraction

Ignores sidebar previews, background tabs, and unrelated content

Works across common CX platforms without configuration

Highlights escalation signals

Customer messages are scanned for escalation indicators: explicit keywords (refund, chargeback, legal), behavioral signals (ALL CAPS, excessive punctuation), and contextual phrases that suggest frustration or intent to escalate. Flagged phrases are gently highlighted inline.

Keyword detection: refund, credit, dispute, complaint, legal, attorney

Behavioral patterns: ALL CAPS, repeated punctuation (!!!), repeated messages

Contextual phrases: 'I already told you', 'this is ridiculous', 'speak to a manager'

Risk level displayed as a small inline badge: Low, Watch, or Escalation

Reviews the agent's draft

As the agent types a reply, the draft text is analyzed in real time. Guardrail checks for language that could create liability — admissions of fault, unauthorized refund promises, guarantees, or overly casual tone. Issues are surfaced inline before the message is sent.

Admissions of fault: 'our mistake', 'we messed up', 'our error'

Unauthorized promises: 'I will refund', 'I can credit you'

Guarantees: 'will be fixed today', 'I guarantee', 'absolutely will'

Configurable response: soft warning or send block, set by managers

Configured by managers

CX managers control the rules. A simple settings interface allows editing keyword lists, toggling between warn and block modes, enabling or disabling supported platforms, and managing a library of approved response snippets that agents can reference.

Edit keyword and phrase lists for detection rules

Toggle between warn mode (advisory) and block mode (prevents send)

Enable or disable Guardrail on specific platforms

Maintain a library of approved response templates

The Full Picture

From inbox to reviewed reply

Guardrail CX fits into the existing workflow. No new windows, no context switching — just quiet, inline guidance where it's needed.

Guardrail CX workflow: inbox to annotation to reviewed draft

Ready to see it working?

Try the interactive demo to experience Guardrail CX on a simulated inbox — or request access to learn more about bringing it to your team.

Runs in the browser. Chrome extension — available soon.