← TKF talk · Slide 14 of 18
Technology stack
🧱 Six components. One machine. No microservices, no orchestration layer, no vendor platforms. Plain tools that have been around for a decade, composed in a new way.
💬 THE CLIENT - OpenWebUI ChatGPT-like browser interface. What the user actually sees. Points at our proxy instead of the cloud.
🎛️ THE PROXY - Perl + Mojolicious Web framework that handles HTTP, streaming, and async I/O. The proxy is built on this.
🧠 THE BRAIN - Brain.pm Custom module. 730 lines of Perl that does chunking, searching, and indexing. The core engine.
🧠 THE BRAIN - SQLite Lightweight database. One file per brain. Stores passages and metadata. No server needed.
🧠 THE BRAIN - PDL Perl Data Language. Does the math that finds the best-matching passages. Fast number crunching.
🤖 THE AI - Ollama + bge-m3 Runs the AI models locally on our hardware. No cloud calls. The thing that reads and writes.
📏 Brain.pm is 730 lines. No Pinecone. No Chroma. No Weaviate. No LangChain. Compare to a typical Python RAG stack: the complexity difference is enormous.