ITI Agents Advanced Features
Advanced Features
Agent Coordination Modes
Navigate to: ITI Agents → Settings → Coordination
Sequential Mode
Agents work one after another in a defined order. Best for projects requiring clear workflow stages.
Parallel Mode
Multiple agents work simultaneously on different aspects. Faster but requires careful coordination.
Hybrid Mode (Recommended)
Intelligent combination of sequential and parallel execution based on task dependencies.
Context Manager
The Context Manager handles runtime context injection:
- Loads comprehensive prompts from database
- Injects project-specific context
- Includes prior agent outputs for coordination
- Manages conversation history
- Tracks agent state across interactions
Performance Tracking
Navigate to: ITI Agents → Analytics
Monitor agent and system performance:
- Agent Usage Statistics – Which agents are used most
- Response Quality Metrics – User satisfaction ratings
- Token Usage – API consumption and costs
- Response Times – Performance monitoring
- Success Rates – Task completion metrics
- Prompt Performance – Which prompt versions work best
Custom Agent Development
Navigate to: ITI Agents → Custom Agents
Create your own specialized agents:
- Define agent purpose and domain
- Write comprehensive prompt (3,500+ words recommended)
- Configure agent capabilities
- Test with sample scenarios
- Deploy to production
Best Practice: Study existing agent prompts before creating custom ones. The comprehensive prompt structure is key to success.
Agent Collaboration
How Agents Work Together
ITI Agents uses sophisticated coordination to ensure agents work effectively together:
- Context Sharing – Agents access outputs from prior agents in the workflow
- Smart Handoffs – System determines which agent should respond based on query content
- Conflict Resolution – When agents have different recommendations, synthesis agents reconcile them
- Iterative Refinement – Agents can revisit and refine their outputs based on new information
Best Practices for Multi-Agent Projects
- Start with Intake Agent to establish comprehensive context
- Let agents work sequentially on complex, interdependent tasks
- Use parallel mode for independent analysis tasks
- Always engage Synthesis agents for holistic recommendations
- Review agent outputs and provide feedback for better results
