Conduit
Book DemoEnroll in Course
  • Start Guide
  • Setup
    • Workspaces
    • Invite and Manage Members
    • Notifications
    • Mobile
    • Flexible Inboxes
      • Create an Inbox
      • Custom Fields
      • Workflow
  • Channels
    • Property Management System
      • Guesty
      • Track
      • Hostaway
      • BookingSync/Smily
      • Lodgify
      • Hostfully
      • OwnerRez
      • Beds24
      • Add and Remove Listings
    • Email
      • Receiving Emails in Conduit
      • Sending Emails from Conduit
      • Alternate Email Addresses
    • WhatsApp
    • Business Phone Number
      • Port Your Phone Numbers to Conduit
        • OpenPhone
        • Aloware
        • Google Voice
        • Dialpad
        • RingCentral
        • Aircall
        • Grasshopper
        • Zoom Phone
    • Chat Widget
    • Airbnb
      • Standalone
      • With a PMS
  • Using The Inbox
    • Inbox Zero
    • AI Messaging
    • Conversation Tags
    • Labels
    • Notes and Mentions
    • Assignment
    • Saved Replies
    • Insights
  • Building Your AI Agents
    • Road to Autopilot
    • Phase 1: Teaching
      • Sources
      • Required Info
      • Teach as You Go
    • Phase 2: Tuning
      • In the Sandbox
      • In the Inbox
      • Advanced Rules
    • Autopilot
  • Workflows
    • Getting Started
  • Create a Workflow
  • Trigger and Action Libraries
  • AI Agent Injections
  • Integrations
    • Slack
    • Zapier
  • Other
    • Conduit Certification
Powered by GitBook
On this page
  • Overview
  • How It Works
  • A Recommended Workflow

Was this helpful?

  1. Building Your AI Agents
  2. Phase 2: Tuning

In the Sandbox

A test environment for you to see and improve your AI's responses

PreviousPhase 2: TuningNextIn the Inbox

Last updated 1 month ago

Was this helpful?

Overview

To trust your AI agent, you need to know how it responds in real-world scenarios. The Sandbox is your testing ground—a space where you can see exactly what your AI knows, identify gaps, and fine-tune its responses.

By simulating guest interactions, you get full visibility into what information your AI is using and where it might fall short. This ensures responses remain consistent, accurate, and aligned with your brand’s standards.

How It Works

  1. Navigate to the Sandbox

    • Select a listing you want to test for

    • Begin simulating guest interactions.

  2. Ask a Question

    • Type a question as if you were a guest.

    • The AI will generate a response based on the knowledge it has for that listing.

  3. Review the AI’s Knowledge Blocks

  • Each response displays the Knowledge Blocks the AI pulled from.

  • Click on a block to review or edit the knowledge behind the response.

  1. Refine & Improve AI Responses

  • If the response isn’t quite right, you can:

    • Edit the Knowledge Blocks used.

    • Add missing knowledge to fill gaps.

    • Remove outdated or inaccurate information.

    These updates will instantly improve how your AI responds in similar situations moving forward.

  1. Test Different Scenarios

  • Impersonate different guest situations (e.g., early check-ins, WiFi issues, lost keys) to see how the AI handles them.

  • Ensure responses remain accurate and consistent across various guest interactions.

A Recommended Workflow

  • If the AI generates an inaccurate or incomplete response in the inbox, jump into Sandbox and try to replicate the situation.

  • Tweak the knowledge until you get the response you want.

  • The more you refine, the more confidently you can move your AI toward Autopilot.

Select the contact type you want to test. This will default to Guest

(see more on different contact types)
Click the gear icon to configure the exact guest situation you want to replicate