Agent Assist

Theme

Quick Start ​

This guide walks you through setting up Agent Assist from scratch. By the end, you will have real-time AI suggestions flowing to your agent desktop and an analytics dashboard tracking performance.

Quick Start Walkthrough

1

Run the Setup Wizard

Configure your integration, knowledge base, LLM, and classifier.

2

Upload Documents

Add content to your knowledge base so agents get relevant suggestions.

3

Generate an Embed Token

Create a secure token to deploy the widget in your CCaaS platform.

4

Test Real-Time Suggestions

Trigger a conversation and verify suggestions appear.

5

Review Analytics

Open the dashboard and confirm data is flowing.

Step 1: Run the Setup Wizard ​

The setup wizard walks you through the initial configuration. Navigate to Setup in the portal sidebar.

The wizard adapts to your assist mode:

Omni Mode (recommended for most deployments):

StepWhat You Configure
IntegrationSelect your CCaaS platform integration (Genesys, Five9, 8x8, Salesforce, Google CCaaS)
Knowledge BaseChoose the default KB and set max sources / confidence threshold
LLM ConfigSelect the LLM integration and system prompt
ClassifierSet the classifier threshold for filtering noise messages
Review & ActivateReview settings and enable the connector

Google Native Mode (for organizations with existing Dialogflow CX):

StepWhat You Configure
IntegrationSelect your CCaaS platform integration
CCAI ProfileLink your Dialogflow Conversation Profile
AudioHookEnable real-time voice transcription
MiddlewareEnable the CCaaS middleware gateway
NotificationsConfigure Pub/Sub notification topics
Review & ActivateReview settings and enable the connector

TIP

Most deployments use omni mode. Choose google_native only if your organization already has a mature Dialogflow CX setup and prefers Google-managed suggestion logic.

After completing the wizard, you can fine-tune all settings under Settings at any time. See the Tenant Configuration guide for details on each setting.

Step 2: Upload Documents ​

Knowledge bases are the source material for agent suggestions. Upload documents, and the RAG pipeline will chunk, embed, and index them for vector search.

Create a Knowledge Base ​

  1. Go to Knowledge > Knowledge Bases (or configure the default KB during the setup wizard)
  2. Click Create Knowledge Base
  3. Give it a name and optional description
  4. Select the embedding model (defaults to your tenant's configured AI model)

Upload Documents ​

Supported formats:

FormatExtensionsNotes
PDF.pdfText-based PDFs; scanned PDFs require OCR
Word.docxPreserves heading structure for better chunking
Plain text.txt, .mdBest for FAQs and structured content
HTML.htmlStripped to text content
CSV.csvEach row treated as a separate chunk
imageKnowledge base document upload interface showing drag-and-drop area and file list
Upload documents to your knowledge base via the Documents tab

Upload files through the Documents tab on your knowledge base.

You can also connect external document sources (SharePoint, Google Drive, S3, Dropbox, OneDrive, Box, Azure Blob) for automatic syncing. See External Document Sources for setup instructions.

TIP

Start with your most frequently referenced documents -- agent scripts, FAQ sheets, and product guides. You can always add more later and the index updates automatically.

Verify Indexing ​

After upload, check the document status in the Documents tab. Each document progresses through: uploaded -> chunking -> embedding -> indexed. Once indexed, the content is available for suggestions.

Step 3: Generate an Embed Token ​

The Agent Assist widget is embedded in your agent desktop via an iframe. It requires a secure embed token to authenticate.

Generate the Token ​

  1. Navigate to Settings in the portal sidebar
  2. Scroll to the Widget Configuration section
  3. Add your CCaaS platform's domain to Allowed Origins (e.g., https://apps.use2.us-gov-pure.cloud for Genesys)
  4. Click Generate Token
  5. Copy the token -- it is only displayed once

Deploy in Your CCaaS Platform ​

Embed the widget using the generated token:

html
<iframe
  src="https://assist.yourdomain.com/?t=YOUR_EMBED_TOKEN"
  width="400"
  height="100%"
  frameborder="0"
  allow="clipboard-write"
></iframe>
imageAgent Assist widget embedded in the Genesys Cloud agent desktop
The Agent Assist widget running inside the agent desktop iframe

The widget automatically:

  1. Validates the embed token via the bootstrap API
  2. Loads the appropriate CCaaS provider adapter (Genesys, Five9, 8x8, Salesforce, Google CCaaS, or standalone)
  3. Handles agent authentication via the provider's OAuth flow
  4. Establishes a Socket.IO connection for real-time suggestions

Provider-specific deployment guides:

ProviderDeployment MethodGuide
Genesys CloudInteraction Widget in Genesys AdminWidget Deployment
SalesforceLightning Web Component in Utility BarSalesforce Integration
Five9Agent Desktop Toolkit iframeWidget Deployment
8x8Partner Integration iframeWidget Deployment
Google CCaaSiframe embedWidget Deployment
StandaloneDirect URL accessWidget Deployment

Step 4: Test Real-Time Suggestions ​

With the tenant configured, knowledge base indexed, and widget deployed, you are ready to test.

  1. Start a test conversation -- Initiate a chat or call through your CCaaS platform
  2. Send a customer message -- Type or speak a question that matches your uploaded knowledge base content
  3. Watch the widget -- Within 2-5 seconds, the widget should display a suggestion with:
imageSuggestion card appearing in the widget with answer text, source citations, and feedback buttons
A real-time suggestion appearing in response to a customer query
  • A generated answer streaming progressively as the LLM generates it
  • Source citations linking back to the original documents with relevance scores
  • Follow-up questions (voice) or quick replies (chat/message)
  1. Provide feedback -- Click the thumbs-up or thumbs-down icon on the suggestion

Troubleshooting ​

If suggestions do not appear:

SymptomCheck
Widget shows "Disconnected"Verify the embed token is valid and the connector service is running
Widget connects but no suggestionsConfirm the knowledge base has indexed documents and the assist_mode is set
Suggestions are irrelevantReview the classifier threshold in Settings and ensure your knowledge base covers the topic
High latency (>10 seconds)Check Service Health for slow components
Only "quick replies" appear, no answerThe classifier is filtering the message as noise -- lower the classifier threshold or add domain keywords

WARNING

Test with real-world customer questions, not generic greetings. The classifier ignores low-signal messages like "hello" or "hi there" to avoid unnecessary RAG queries.

Step 5: Review Analytics ​

Open the Dashboard in the Agent Assist portal to confirm that data is flowing correctly.

Key Dashboards ​

DashboardWhat It ShowsLocation
DashboardActive conversations, search rate, latency, acceptance rateHome page
Feedback AnalyticsThumbs-up/down rates, reason codes, comment trendsSupervisor > Reports
RAG PerformanceRetrieval latency, hit rate, confidence distributionSupervisor > Reports
Cost & UsageToken usage, cost breakdown, per-agent attributionSupervisor > Reports
Service HealthPub/Sub delivery, latency, error ratesAdministration > Analytics

After your test conversation, verify that:

  • The conversation appears in Conversations > History
  • Feedback events show in the Feedback Analytics dashboard
  • RAG queries are logged with latency and confidence metrics

TIP

It may take up to 60 seconds for analytics data to appear after a conversation ends. The reporting pipeline processes events in near-real-time batches.

Next Steps ​

Last updated:

OmniBots Agent Assist

Copyright 2026 OmniBots. All rights reserved.