•1 min read
Building a Local LLM Gateway on a Mac Mini
A small OpenAI-compatible proxy in front of Ollama — Bearer auth, model routing, streaming — so no cluster pod has to hold a vendor API key.
Every post I've published, newest first.
A small OpenAI-compatible proxy in front of Ollama — Bearer auth, model routing, streaming — so no cluster pod has to hold a vendor API key.
Three onboarding gaps closed in one March — friends couldn't join without an account, mobile re-rendered on every scroll, and the UI had drifted from its own wireframes.
Multiplayer games forced a staging branch and a UI for trust — FluxCD kept writing because ArgoCD cannot decrypt SOPS.
Adding a new board game stopped being a backend project the day the server stopped importing game engines — YINSH proved it in a weekend.
Four weeks took a raw k3s box to 14 Knative functions, FluxCD GitOps, and sub-second cold starts — the layering order that made it work.