Agentic AI

Control Expedify with natural language using our AI-powered assistant. Let AI agents autonomously execute multi-step tasks.

What is Agentic AI?

Agentic AI is an intelligent assistant that understands natural language and executes actions across Expedify. Instead of clicking through menus, simply tell it what you want:

"Find all contacts from Acme Corp who haven't been contacted in 30 days
and create follow-up tasks for each"

Agentic mode allows an AI to:

• Understand complex requests
• Break down into steps
• Execute tools/skills
• Adapt based on results
• Complete the task

Instead of fixed workflows, the AI decides what to do.

How It Works

┌─────────────────────────────────────────────────────────────┐
│                      AGENTIC AI                              │
├─────────────────────────────────────────────────────────────┤
│                                                              │
│  User: "Create a summary for John Smith"                    │
│                    │                                         │
│                    ▼                                         │
│  ┌─────────────────────────────────────────────────────┐   │
│  │              ORCHESTRATOR AGENT                      │   │
│  │                                                      │   │
│  │  1. Parse intent → CRM query + AI generation        │   │
│  │  2. Plan steps  → Search contact, Generate summary  │   │
│  │  3. Select skills → crm_search, llm_generate        │   │
│  └─────────────────────────────────────────────────────┘   │
│                    │                                         │
│                    ▼                                         │
│  ┌─────────────────────────────────────────────────────┐   │
│  │                 SKILL EXECUTION                      │   │
│  │                                                      │   │
│  │  [crm_search]     →  Found: John Smith, Acme Corp   │   │
│  │  [llm_generate]   →  Generated summary: ...          │   │
│  │  [crm_update]     →  Saved summary to contact        │   │
│  └─────────────────────────────────────────────────────┘   │
│                    │                                         │
│                    ▼                                         │
│  Response: "I've generated and saved a summary for          │
│            John Smith. Here's what I created: ..."          │
│                                                              │
└─────────────────────────────────────────────────────────────┘

A simplified view of the orchestration loop:

User Request
    ↓
AI Agent (Orchestrator)
    ↓
Thinks → Plans → Executes
    ↓         ↓
[Skill 1] [Skill 2] [Skill 3]
    ↓         ↓         ↓
   Results combined
    ↓
Final Response

Core Components

Orchestrator Agent

The brain that:

• Understands your intent
• Breaks down complex requests
• Selects appropriate skills
• Coordinates execution
• Synthesizes responses

Skills

Pre-built capabilities (actions) the agent can use:

Category	Skills	Description
CRM	search, create, update, delete	Manage CRM records
Workflows	execute, status, list	Run and monitor workflows
Knowledge Base	search, add	Query and update KB
Analytics	query, report	Generate insights
Channel	email, sms	Send messages
Data	transform, aggregate	Process data

Memory

Context the agent maintains:

• Conversation history
• User preferences
• Recent actions
• Entity context

Enabling Agentic Mode

In Chat

Access via the Agentix chat panel:

1. Open Expedify chat (click the chat icon in the navigation)
2. Toggle Agentic Mode on
3. Type your request naturally and ask complex questions
4. Watch as the agent thinks and acts, executing skills automatically
5. Review results and continue the conversation

In Workflows

Use the Agentic Node:

1. Add Agentic node
2. Configure available skills
3. Set the task prompt
4. Agent orchestrates execution

Using Agentic AI

Natural Language Examples

CRM Operations:

"Find all deals worth more than $50,000"
"Create a contact for Jane Doe at jane@example.com"
"Update the status of deal #123 to 'Negotiation'"
"Show me tasks due this week"

Workflow Operations:

"Run the lead scoring workflow for contact ABC"
"What's the status of the onboarding workflow?"
"List all active workflows"

Knowledge Base:

"Search our knowledge base for refund policy"
"What does our documentation say about API limits?"

Analytics:

"How many deals did we close last month?"
"What's our current pipeline value?"
"Show me contacts by lead source"

Multi-Step Operations:

"Find all contacts who opened our last email but didn't click,
then create a follow-up task for each assigned to me"

"Generate a weekly report of new contacts,
summarize the trends, and email it to the sales team"

Agent Behavior

Thinking Process / Thought Bubbles

The agent shows its reasoning in real-time:

• Understanding the request
• Planning approach
• Executing steps
• Evaluating results

┌─────────────────────────────────────────────────────────────┐
│ 💭 Thinking...                                               │
│                                                              │
│ • Understanding request: Find inactive contacts              │
│ • Planning: Need to query contacts, filter by last_activity │
│ • Executing: crm_search skill                                │
│ • Found 23 contacts matching criteria                        │
│ • Generating response                                        │
└─────────────────────────────────────────────────────────────┘

Tool Calling

When the agent needs data or actions:

Agent: I need to find deals closing this month.
       Let me search the CRM...

[Executes: crm_search(entity="deals", filter="close_date:this_month")]

Agent: Found 5 deals. Let me analyze them...

Skill Cards

See skill execution details:

┌─────────────────────────────────────────────────────────────┐
│ 🔧 crm_search                                      ✓ Success │
├─────────────────────────────────────────────────────────────┤
│ Input:                                                       │
│   entity: contacts                                           │
│   filters:                                                   │
│     - field: last_activity                                   │
│       operator: older_than                                   │
│       value: 30 days                                         │
│                                                              │
│ Output:                                                      │
│   count: 23                                                  │
│   results: [Contact 1, Contact 2, ...]                      │
│                                                              │
│ Duration: 245ms                                              │
└─────────────────────────────────────────────────────────────┘

Error Handling

If a skill fails:

• Agent acknowledges error
• Tries alternative approach
• Reports if blocked

Use Cases

Research Assistant

"Find all contacts at enterprise companies who visited our pricing page last month"

Agent:

1. Searches visitors by page
2. Filters by company size
3. Matches to contacts
4. Returns list

Data Analysis

"What's our best-performing campaign this quarter?"

Agent:

1. Queries campaign data
2. Calculates metrics
3. Compares performance
4. Provides analysis

Task Automation

"Create a follow-up task for every deal that hasn't been updated in 30 days"

Agent:

1. Finds stale deals
2. Creates tasks for each
3. Reports completion

Best Practices

Be Specific

# Good - Clear and specific
"Find contacts from Acme Corp created in the last 7 days"

# Vague - Requires clarification
"Find some contacts"

Provide Context

# Good - Includes context
"For the deal we discussed earlier, update the stage to 'Proposal Sent'"

# Missing context - Which deal?
"Update the deal stage"

Use Confirmations

For important actions, the agent will ask for confirmation:

Agent: "I found 150 contacts matching your criteria.
        Creating tasks for all of them will generate 150 tasks.
        Should I proceed?"

User: "Yes, go ahead"

Summary

1. Clear requests - Specific instructions
2. Start small - Simple tasks first
3. Review executions - Check agent behavior
4. Set guardrails - Prevent unintended actions
5. Provide context - Help agent understand

Configuration

Enable/Disable Skills

Control which skills the agent can use:

1. Navigate to Settings → Agentic AI
2. Toggle skills on/off
3. Set skill-specific permissions

You can also manage skills from Automation > Agent Skills:

1. Go to Automation > Agent Skills
2. Enable/disable skills
3. Configure permissions
4. Set rate limits

Customize Behavior

Adjust agent behavior:

Setting	Description	Default
Confirmation Required	Require confirmation for destructive actions	Yes
Max Results	Limit results returned	100
Verbose Mode	Show/hide detailed thinking	Off
Auto Execute	Execute immediately vs. show plan first	Off

System Prompt

Define agent behavior:

You are a helpful CRM assistant for Expedify.
You can search and update CRM records, analyze data,
and help users accomplish their tasks efficiently.

When searching, always confirm the scope before executing.
When creating records, summarize what you'll create first.

Guardrails

Set limitations:

• Maximum steps per request
• Confirmation for destructive actions
• Scope restrictions

Monitoring

Execution Logs

View agent activity:

• Steps taken
• Skills executed
• Time per step
• Results

Analytics

Track agent performance:

• Success rate
• Average steps
• Common requests
• Failure reasons

API Access

Use agentic AI programmatically:

POST /api/agentic/chat
Content-Type: application/json

{
  "message": "Find all deals closing this month",
  "context": {
    "conversation_id": "conv_123"
  }
}

Response (SSE Stream):

event: thinking
data: {"thought": "Parsing request..."}

event: skill_start
data: {"skill": "crm_search", "input": {...}}

event: skill_end
data: {"skill": "crm_search", "output": {...}}

event: response
data: {"message": "I found 12 deals closing this month..."}

Troubleshooting

Issue	Solution
Agent stuck	Check skill availability
Wrong results	Improve request clarity
Too many steps	Simplify the task
Permission denied	Check skill permissions

Related & Next Steps

• Knowledge Base - AI-searchable documents
• Agent Skills
• Chat Widget - Embed AI chat in your app
• Workflows - Automate with visual workflows