MODULE 1  ·  DAY 1

Container-Native GenAI

Why the open standard — not Docker — is the through-line for AI

Gourav Shah  ·  School of DevOps & AI  ·  Lesson + Lab + Quiz

What you'll learn

Five ideas that shape every lab in the rest of the course.

Concepts first — the lab then proves the container-to-native-model wiring with your own hands.

The assumption that just broke

Docker Desktop is now paid for orgs over 250 staff or $10M revenue. "Container = Docker" no longer holds.

The standard under Docker was always open. This module builds on that standard, not on any one vendor.

Container-native, not Docker-native

An OCI image is a shipping container — the same steel box loads onto any carrier. Docker is just one carrier.

One file, four runtimes, same result — the carrier you pick is your business, not the course's.

What containers buy an AI stack

Think of each component as a hermetically sealed shipment you can open on any machine.

Package, serve, isolate, ship — the four jobs a container does for every piece of your AI stack.

The Apple-Silicon GPU reality

macOS containers are like guest rooms with no power outlets — the building has power, the rooms aren't wired in.

The single most important practical lesson: get this wrong and every lab crawls. So don't fight it — wire around it.

The fix: native server, containerized everything else

Serve the model natively on Metal; put the app, agent and vector DB in containers; bridge them with one magic hostname.

On Windows + WSL2 + NVIDIA the toolkit passes the GPU in, so the server can be containerized there.

One universal contract: the OpenAI-compatible endpoint

The API is a wall socket. Swap the power station behind it and your appliance never notices.

The same request runs against Ollama on a laptop, vLLM on a GPU VM, or OpenAI — every lab speaks this language.

The 2026 map: declarative vs orchestration

Two ways to build an agent. Start with the light one; reach for the heavy one only when you must.

The rule: start declarative; add orchestration only when the task has hard sequencing you can't express in tools.

Anatomy of a declarative agent

Four plain parts define an agent — no framework code required to describe who it is.

Identity, skills, tools, guardrails — all in files you can read, diff, and version like any other code.

Meet Acme — two connected tools

One fictional company, runbooks nobody reads, and two AI tools that fix it across the two days.

Use Case B uses Use Case A as one of its tools — skills compound instead of tangling.

The build ladder — one step per module

Every module adds exactly one rung to the same growing compose.yaml.

Module 1 is step 0: a runtime and a model responding to a call. You hand-author every block from here.

TO THE LAB

Build it once. Run it anywhere.

In the lab you'll prove this wiring yourself — container calls native model, and it answers.

Now open Module 1 · Lab.  ·  Gourav Shah · School of DevOps & AI