An agent is a configurable, goal-driven AI component that understands user input, reasons using context, and takes action by invoking tools or delegating tasks to other agents.
Each agent has four core components:
| Component | Purpose |
|---|
| Scope | Defines what tasks the agent is responsible for |
| Instructions | Governs how the agent behaves and responds |
| Tools | Enables the agent to take actions and retrieve data |
| Knowledge | Provides context for accurate, grounded responses |
╭────────────────────────────╮ ─┐
│ Agent Name │ │ Used by the
╰────────────────────────────╯ ├─ Orchestrator to
│ │ select the agent
╭────────────────────────────╮ │
│ Agent Description │ │
╰────────────────────────────╯ ─┘
│
╭────────────────────────────╮ ─┐
│ Agent Definition │ │ Help in defining the
│ · Scope │ ├─ capabilities of the
│ · Instructions │ │ agent to reason,
╰────────────────────────────╯ │ plan and execute tasks
│ │
╭────────────────────────────╮ │
│ Agent Tools │ │
╰────────────────────────────╯ ─┘
Before You Begin
- Create an Agentic App. Agents must be associated with an app and cannot exist independently. See Create an Agentic App.
- Configure an AI model with tool-calling support. The Platform supports OpenAI, Gemini, Anthropic, and Azure OpenAI. See Add an External Model.
- Verify your role. You need App Owner or App Developer access. Viewers cannot create agents.
Choosing How to Create an Agent
| Option | When to use | What you configure |
|---|
| From Scratch | You want full control over all settings | Profile, instructions, tools, and knowledge |
| Connect External Agent | You have an agent hosted elsewhere | Connection details only |
| From Marketplace | You want a preconfigured, ready-to-use agent | Minimal edits required |
Step 1: Define the Agent Profile
The agent profile establishes the agent’s identity and purpose. The orchestrator uses it to identify the most suitable agent for each incoming task.
Name
Choose a descriptive name that reflects the agent’s function.
Good Avoid
──── ─────
Order Support Agent Agent 1
Billing Assistant Helper
Travel Booking Specialist Bot
Description
Write a description that tells the orchestrator when to route requests to this agent. Be specific about capabilities.
Good: Handles customer inquiries about order status, shipping updates,
delivery modifications, and order cancellations. Can look up
orders by order number, email, or phone number.
Weak: Helps with orders.
AI Model
Select the model that powers this agent’s reasoning.
| Use case | Recommended model |
|---|
| Complex reasoning | GPT-4, Claude 3 Opus |
| Fast responses | GPT-3.5-turbo, Claude 3 Haiku |
| Code generation | GPT-4, specialized code models |
| Cost-sensitive tasks | Smaller models |
Context Window
Set how many conversation messages the agent retains. Higher limits provide more context but increase token usage.
| Messages | Use case |
|---|
| 25 | Simple Q&A, transactional tasks |
| 50 | Standard conversations (default) |
| 100 | Multi-step processes |
| 200 | Complex, long-running workflows |
Step 2: Configure Agent Definition
The agent definition is the system prompt. It describes the agent’s role, the tasks it can perform, and the rules it must follow. Well-structured instructions ensure the agent responds consistently and follows expected workflows.
Structure
## Role
Who is this agent and what is their purpose?
## Guidelines
How should the agent behave?
## Response Format
How should responses be structured?
## Constraints
What should the agent never do?
## Examples (optional)
Concrete examples of good responses.
Writing Effective Instructions
Define a specific role:
# Too vague
You are a helpful assistant.
# Better
You are a product specialist who helps customers find the right
products based on their needs and budget.
### Out of Scope
- Payment processing and refunds (→ Billing Agent)
- Product recommendations (→ Sales Agent)
- Technical product support (→ Tech Support Agent)
## Constraints
Never:
- Share or confirm full credit card numbers
- Make promises about timelines you can't guarantee
- Engage with abusive or threatening language
- Make up information when you don't know the answer
## Response Format
- Simple questions: answer in 1-2 sentences
- Process explanations: use numbered steps
- Multiple options: use bullet points
Complete Example
## Role
You are a customer support agent for CloudStore, an online electronics
retailer. Help customers with orders, returns, and product questions
with a friendly, efficient approach.
## Guidelines
### Communication
- Be concise—aim for 2-3 sentences when possible
- Use a warm but professional tone
- Match technical depth to the customer's level
### Order Inquiries
- Always look up the order first
- Provide status, tracking, and expected delivery
- Proactively mention any delays or issues
### Returns and Refunds
- Explain the policy clearly (30 days, original condition)
- Guide through the return process step-by-step
- Offer exchanges as an alternative when appropriate
## Response Format
- Simple questions: 1-2 sentence answers
- Process explanations: numbered steps
- Multiple options: bullet points
## Constraints
Never:
- Share full payment details
- Promise delivery dates not confirmed by tracking
- Process refunds over $500 without supervisor approval
## Escalation Triggers
Transfer to a human agent when:
- Customer explicitly requests it
- Issue involves disputes over $500
- Customer expresses significant frustration
Step 3: Add Tools
Tools are the functional extensions of an agent’s intelligence. They enable agents to fetch data from external systems, create or update records, and trigger workflows or APIs.
Agent Platform supports three tool types:
- Workflow Tools — Trigger automated workflows
- Code Tools — Execute custom logic
- MCP Tools — Connect via the Model Context Protocol
Adding a Tool
- In agent configuration, go to Tools → + Add Tool
- Choose Create new to build a tool, or Import existing to reuse one from the library
- Configure the tool with a clear name, description, and parameter definitions
The description is used by the LLM to decide when and how to call the tool—write it carefully.
name: get_order_status
description: |
Retrieves the current status of a customer order including
shipping details, estimated delivery, and tracking number.
parameters:
order_id:
type: string
description: The unique order identifier (for example, ORD-12345)
required: true
Step 4: Add Knowledge
Link knowledge sources to enable RAG-based responses. The Platform uses the Search AI application to search across enterprise knowledge bases, document repositories, FAQs, and external systems. You can connect one or more Search AI applications from the same workspace.
Connecting Knowledge
- In agent configuration, go to Knowledge → + Connect Knowledge
- Select an existing Search AI application or create a new one
Supported Sources
| Source | Best for |
|---|
| Documents | Policies, guides, FAQs |
| Web crawler | Website content, help centers |
| Confluence | Internal documentation |
| SharePoint | Enterprise content |
| Custom API | Dynamic data sources |
Step 5: Configure Delegation Logic
Applies to: Apps using the Adaptive Network orchestration pattern only.
Delegation rules define how this agent routes or hands off a user query to another agent within the application. These rules determine which agent is best suited to handle specific intents, tasks, or contexts.
If the app uses any other orchestration pattern, delegation configuration isn’t required and won’t be available in the UI.
Test and Deploy
Test the Agent
- Click Test in the agent toolbar
- Run conversations covering:
| Scenario | What to verify |
|---|
| Happy path | Standard requests the agent handles correctly |
| Edge cases | Invalid inputs, missing information, ambiguous requests |
| Boundary cases | Requests outside the agent’s defined scope |
| Error cases | Tool failures, escalation triggers |
Deploy
- Click Save to preserve your configuration
- Run Diagnostics to validate all dependencies are properly configured
- Publish the app to make the agent live
Agent Types by Use Case
| Type | Scope | Tools | Knowledge |
|---|
| Customer Service | Order tracking, returns, product questions | CRM lookup, order management, ticketing | FAQs, product docs, policies |
| Sales | Product recommendations, pricing, checkout | Product search, cart management, payment | Catalog, pricing, promotions |
| HR / Employee | Leave requests, benefits, IT support | HR systems, ticket creation, knowledge search | HR policies, IT guides, company wiki |
| Process | Document processing, approvals, data entry | Document extraction, workflow triggers, database updates | Process docs, compliance rules |
