Architecture Overview

This is the current-state architecture of the razizu platform: a single-machine cluster running a small internal agent fleet and three daily AI-narrated podcast shows, gated by a single shared login. There is no shared AI gateway anymore — every app that needs a model calls its outside provider directly; see LLM Gateway.

This describes what’s meant to be running, not necessarily what’s live on the machine at this exact instant. Treat any single detail as a snapshot.

The whole fleet, one diagram

Diagram
Langfuse (top right) is still deployed but currently has no incoming edge in this diagram on purpose — nothing in the fleet feeds it new traces since the shared AI gateway that used to forward calls to it was removed. See Langfuse.

Reading order

  • How to Read These Docs explains the split between this component (structure) and Use Cases (outcomes).

  • Platform → VPS & K3s onward walks the shared infrastructure layer: the machine itself, ingress and TLS, authentication, app layout, how apps reach AI providers, observability, storage, and CI/CD.

  • Podcast Agent → Overview onward covers the one application in this fleet complex enough to warrant its own module — three daily-generated news podcast shows from one service.

What’s here that older docs didn’t cover

This is a living picture of a fleet that’s grown over time — it now includes a small internal agent fleet, three daily-generated podcast shows with public feeds, and a log-search stack, none of which existed when this platform first started. This documentation site itself is part of that fleet, sitting behind the same shared login as everything else internal. It has also shed a component since: the shared AI gateway every app used to route through was removed in favor of direct provider access — see LLM Gateway.