Modern agentic AI applications follow various architectural patterns. This guide shows which patterns each framework supports and how Monocle instruments them.
- Single Agent - One agent handles all tasks
- Multi-Agent Sequential - Agents work in sequence, passing results
- Multi-Agent Parallel - Agents work simultaneously
- Orchestrator/Supervisor - Coordinator agent manages specialized sub-agents
- Agent Delegation/Handoffs - Agents transfer control to other agents (different frameworks use different terms:
handoffs,handoff_to_*,transfer_to_*) - Session/Thread - Multi-turn conversation context preservation
| Framework | Single | Sequential | Parallel | Orchestrator | Delegation/Handoffs | Session/Thread | Monocle Support |
|---|---|---|---|---|---|---|---|
| Google ADK | ✅ | ✅ SequentialAgent |
✅ ParallelAgent |
✅ LoopAgent |
✅ Built-in |
✅session_id |
Full ✅ |
| LangGraph | ✅ | ✅ | ✅ | ✅ Supervisor |
✅transfer_to_* |
✅thread_id |
Full ✅ |
| OpenAI Agents | ✅ | ✅ | ✅ | ✅handoffs=[] |
✅thread |
Full ✅ | |
| Microsoft Agent Framework | ✅ | ✅ | ✅ | ✅handoff_to_* |
✅thread |
Full ✅ | |
| LlamaIndex | ✅ | ✅ | ✅ | ✅session_id |
Partial |
||
| CrewAI | ✅ | ✅ Sequential |
✅ Hierarchical |
✅ | ✅ | Partial |
|
| Strands | ✅ | ✅ | ✅session_id |
Partial |
|||
| LangChain | ✅ | ✅ | ✅ | Basic |
|||
| Haystack | ✅ | ✅ Pipeline |
✅ | Basic |
Legend: ✅ Full support |