Agentic System
Design and configure your multi-agent AI system with the visual agent builder.
The Agentic System is where you design, configure, and refine your multi-agent system. The AI architect generates an initial system based on your use case, and you can modify it through natural language conversation or direct configuration.
How It Works
When you enter Control Hub, the system automatically generates a multi-agent system tailored to your specific use case. The generated system includes:
- Specialized Agents -- Custom agents designed specifically for your use case, each with defined roles, tools, and capabilities
- Platform Agents -- Three dedicated agents that manage overall infrastructure, security, and observability across your system
You can refine this system in two ways:
-
Visual Configuration -- Expand any agent card to directly modify its model, tools, memory, and system prompt.
-
Natural Language -- Use the chat interface to request changes like "Add a sentiment analysis agent" or "Give the coordinator long-term memory."
Agent Configuration
Each agent in your system can be configured with:
Model Selection
| Model | Best For |
|---|---|
| Claude Opus 4.5 | Complex reasoning, multi-step analysis, high-stakes decisions |
| Claude Sonnet 4.5 | Balanced capability and cost, general-purpose tasks |
| Claude Haiku 4.5 | Fast responses, simple routing, lightweight operations |
Tools
Connect agents to external capabilities:
- RAG (Knowledge Base) -- Search your company documents, wikis, and data sources
- MCP (External Integrations) -- Connect to business systems like Slack, Gmail, Jira, Salesforce
- Web Search -- Real-time internet search for current information
Memory
| Type | Purpose |
|---|---|
| Short-term (SEMANTIC) | Maintains context within a conversation session |
| Long-term (SUMMARY) | Persists important information across sessions |
| Semantic (USER_PREFERENCE) | Learns and remembers user preferences over time |
System Prompt
The instructions that define the agent's behavior, personality, expertise, and constraints. This is where you shape how the agent thinks and responds.
Multi-Agent Orchestration
The generated system includes orchestration patterns that define how agents communicate and collaborate:
- Specialized agents handle specific domains or tasks based on your use case
- Platform agents manage infrastructure, security, and observability
- Handoff patterns enable seamless transitions between agents
Approving Your Agentic System
When you're satisfied with the configuration, click "Approve" to trigger deployment. Once deployment completes, you are taken to the Test Bed for feature generation and preview of your deployed application.