Orchestration¶
A six-layer orchestration stack that moves agents from single-task executors to autonomous operators.
Layer 1: Hermes Agent → Execution kernel (tool loop, skills, session mgmt)
Layer 2: CrewAI → Multi-agent coordination (delegate, parallel work)
Layer 3: LangGraph → Stateful workflows (checkpoints, conditional routing)
Layer 4: Reflexion → Self-improvement loops (evaluate, reflect, retry)
Layer 5: Cron + Scheduling → Persistent operations (24 recurring jobs)
Layer 6: Governance → Safety rules, monitoring, drift detection
Pages¶
| Framework | What It Does | Doc |
|---|---|---|
| Hermes Agent | Execution kernel — tool loop, skills loading, session management, model routing | Read → |
| CrewAI | Multi-agent coordination — task delegation, parallel workstreams, agent roles | Read → |
| LangGraph | Stateful graph workflows — checkpoints, conditional routing, subgraph isolation | Read → |
| Reflexion | Self-improving loops — evaluation, reflection, memory, iterative improvement | Read → |
Why This Stack¶
Single-agent chat patterns break at scale. Agents need: - Memory: What happened last session? (GBrain, Honcho) - Coordination: Who does what? (CrewAI, LangGraph) - Improvement: Did that work? (Reflexion) - Persistence: Run without me watching? (Cron, governance)
This stack answers all four.
← Knowledge | MCP Integration → | ↑ Home