HOW I BUILT IT
The architecture is built on three core layers.
The total cost for individual use runs around $50/month in API and infrastructure costs. The point of this project was never just to save myself time. It was to build the same RAG + agentic architecture I recommend to enterprise clients, prove it works at the personal scale first, and use it as the foundation for the consulting work I do. I eat my own dog food. That's how I know it works.
🔌 Current integrations include Outlook, ClickUp, Fathom, and Google Drive. Phase 1 is a personal pilot. Phase 2 is team deployment.
n8n
Handles orchestration. It's the system's nervous system: listening for triggers across connected platforms, routing information to the right workflow, and executing tasks automatically. Every automation, every briefing, every meeting summary runs through n8n.
Weaviate
Handles memory. It's a vector database that stores and retrieves information semantically, meaning the system can find the right context even if you don't phrase the query exactly right. Client data is segmented by ID so information about one project never informs an output for another.
Interchangeable Frontier Models
Handle intelligence. Accessed via OpenRouter, these models do the reasoning, summarization, and drafting. Every AI-generated output is logged and labeled: either [Verified] (sourced from actual documentation) or [Inference] (AI-generated). You always know when to double-check.