AI Workdeck Self-Hosted Architecture combines a desktop Electron/Vue client with a Spring Boot backend orchestrating RAG pipelines, document AI, and LLM inference entirely on-premises. The system routes user requests through a Spring Boot API to an Agent Orchestration Engine that coordinates plugin execution, document editing via WPS, and AI service calls including embedding, OCR, and LLM endpoints. Evidence-Chain Store captures audit trails while Vector Database, Prompt Cache, and Model Registry enable efficient retrieval-augmented generation with safety guardrails. This architecture empowers organizations to deploy enterprise AI document workflows with full data sovereignty, compliance control, and customizable model selection. Fork and customize this diagram on Diagrams.so to adapt component choices, add external integrations, or document your own self-hosted AI stack. The modular plugin system and message queue design support scaling inference workloads independently from API services.