Command Center - Hermes w/ Ollama & Lemonade Stack.
My AI stack is becoming an organization, not a chatbot
I’m building around Hermes as the Chief of Staff, not trying to make one giant model do everything.
Hermes owns intent, delegation, memory, projects and orchestration. Under it are specialist swarms for engineering, product, marketing/social, sales, research, operations and customer support.
Models are interchangeable compute. Local inference handles routine work; stronger local models handle coding/reasoning; frontier models are escalation. A model director chooses based on capability, cost, latency and availability instead of hard-wiring the organization to one LLM.
Tools sit underneath the agents: terminal/SSH, Git, browser/search, email, social, payments, communications, automation and infrastructure.
The important part is state lives above the chat. Identity, memory, project truth, decisions, tasks and handoffs persist independently. A new session should be another employee walking into the same office—not somebody waking up with amnesia.
Kanban becomes the observable work layer: Intake → Triage → Ready → Running → Blocked → Review → Done. Every card represents real agent work and can be opened to see exactly what the agent is doing.
Phone = command center.
Desktop = deep work.
Smart glasses = ambient interface: voice in, context/alerts/approvals/results out.
Sessions are temporary. Models are replaceable. Devices are interfaces. The organization is permanent.
That’s the architecture.