An Agentic App is the top-level container that groups agents, tools, and knowledge bases into a deployable application.
Prerequisites
- Access to Agent Platform with appropriate permissions to create and manage apps.
- An AI model with tool-calling support added to the Platform. Supported providers: OpenAI, Gemini, Anthropic, Azure OpenAI.
See Add an External Model.
App Development Lifecycle
┌──────────────────┐ ┌─────────────────┐ ┌─────────────────┐ ┌─────────────────┐
│ │ │ │ │ │ │ │
│ 1. Create an App │────▶│ 2. Configure │────▶│ 3. Test │────▶│ 4. Deploy │
│ │ │ Agents │ │ │ │ │
└──────────────────┘ └─────────────────┘ └─────────────────┘ └─────────────────┘
│ Set name, icon, Define roles, Run diagnostics Publish the app
│ description, instructions, and simulate for end users.
│ and AI model. tools, and scenarios.
│ knowledge sources.
Steps
1. Create an App
-
Sign in to the Agent Platform and go to Autonomous Agents.
-
On the Agentic Apps page, click +New App.
-
Fill in the app details:
| Field | Description |
|---|
| Name | A name that reflects the app’s core function. |
| Icon | An icon to identify the app. |
| Description | Describes the app’s purpose and scope. Used to constrain the app to relevant queries and prevent responses based on pre-trained knowledge. |
| AI Model | The model and connection the app will use. |
{
"name": "Customer Support Assistant",
"description": "Handles billing inquiries, order tracking, and returns. Responds only to support-related queries.",
"ai_model": {
"provider": "anthropic",
"connection": "claude-sonnet-4-6"
}
}
-
Click Create App.
The app overview loads automatically. It summarizes:
| Section | What it shows |
|---|
| App Summary | Selected LLM, orchestration pattern, and an Edit option. |
| Agents | Autonomous agents linked to the app, with names, descriptions, and linked tools. |
| Tools | All tools available to agents in the app. |
| Knowledge | Knowledge bases linked to the app. |
App Profile
Click the Edit icon on the app’s Overview page to access the App Profile. Review and update these settings as needed.
| Field | Description |
|---|
| Name | Display name given to the app. |
| Description | A brief summary outlining the app’s purpose. |
| Context Window Limit | Defines the maximum number of messages the application retains in the conversation context. Uses a rolling window approach — once the limit is reached, the oldest messages are removed. Minimum: 25, Maximum: 200. Consider the model’s token limits when setting this field. |
| External Agent | Enable to allow the application to communicate with external agents through proxy agents within the app. Learn more. |
App Orchestrator
The App Orchestrator page lets you select and configure the orchestration pattern for the app. An orchestration pattern defines how multiple agents coordinate, communicate, and execute tasks — including how they delegate work, control execution flow, and resolve conflicts.
The Platform supports the following orchestration patterns:
| Pattern | Description |
|---|
| Single Agent | A single agent independently handles requests and generates responses. Ideal for apps with one primary capability. Learn more. |
| Supervisor | A central controller analyzes requests, delegates tasks to specialized agents, and synthesizes responses. Learn more. |
| Adaptive Network | Agents collaborate in a distributed manner, intelligently routing tasks based on capabilities and context. Learn more. |
Orchestrator Configuration
For each orchestration pattern, configure the following:
| Field | Description |
|---|
| Default AI Model | The AI model the app uses for all operations. Use the gear icon to configure model settings. Configuration properties vary by model — refer to the model’s official documentation for details. |
| Voice-to-Voice Interactions | Enable to allow users to interact with the app via real-time voice conversations. Once enabled, select the AI model for speech processing and voice responses. See Supported Models. Use the gear icon to modify model settings. |
| Behavioral Instructions | Guidelines for the agent’s behavior. The system automatically adds these to the orchestrator prompt and the system prompt of each agent. Click Modify Instructions to enter the prompt. |
Single Agent Configuration
All user requests are routed directly to the agent. Since no supervisor is involved, the agent’s prompt serves as the primary instruction set for the underlying model.
The Platform constructs a consolidated prompt by combining the following components in order:
- Agent Prompt — Core instructions that define the agent’s role and behavior.
- Behavioral Instructions — Guidelines that control tone, constraints, and response style.
- Tools Assigned to the Agent — Tool definitions available for the agent to invoke.
- Events Enabled in the Application — Event-related context.
Learn more.
Supervisor Configuration
In addition to the common configuration above, configure the following for the Supervisor pattern:
| Field | Description |
|---|
| Orchestrator Prompt | Instructions for the supervisor, including decision-making guidelines for the orchestrator. |
| Orchestration Prompt for Voice-to-Voice Interactions | Instructions for the supervisor specifically for voice interactions. |
Adaptive Network Configuration
In addition to the common configuration above, configure the following for the Adaptive Network pattern:
| Field | Description |
|---|
| Initial Agent | The agent that serves as the first point of contact for each task. This agent receives the user’s request, processes the initial requirements, and begins task execution. |
Adaptive Network doesn’t support realtime models.
2. Configure Agents
Define agents with specific roles, instructions, tools, and knowledge sources.
See Set Up an Agent.
3. Test
Run Diagnostics before testing. Diagnostics validates agents, tools, events, and configurations against production-readiness criteria and flags issues before deployment.
Then simulate scenarios to evaluate agent behavior and responses.
See Simulate and Test the App.
4. Deploy
Publish the app for users. See App Deployment 