Stack map: n8n, Open WebUI, and models

• Understand the role of n8n, Open WebUI, Ollama, vLLM, and a custom API.
• Separate chat, automation, model, and data so you can maintain the system.
• Design a first version that is cheap, private, and scalable.

users
  -> Open WebUI: internal chat, models, tools
  -> n8n: workflows, webhooks, approvals, integrations
  -> internal API: business rules and permissions
  -> model:
       local: Ollama / llama.cpp
       GPU production: vLLM
       occasional cloud: external provider
  -> data:
       Postgres / Qdrant / files
  -> operations:
       logs, backups, alerts, limits

Decision rule

• Open WebUI: private interface for users, models, tools, and knowledge.
• n8n: visible automations, webhooks, connectors, approvals, and business logs.
• Ollama: simple local entry point for models on modest CPU/GPU.
• vLLM: model serving with GPU and higher concurrency.
• Custom API: permissions, sensitive rules, and operations you don't want left in an editable workflow.