~/setup
One command sets up everything the hands-on track of this course needs. It opens a guided walk-through in your terminal: it checks what your machine already has, explains each thing it wants to install, and asks before touching anything. By the end you have a real agent running on your own hardware, and the skills taught here produce agents you can deploy inside anything you build.
curl -fsSL https://lyceumagents.com/setup.sh | bashPrefer to read code before you run it? Good instinct — that habit is taught in module 2.06. Download it first, look it over, then run it:
curl -fsSL https://lyceumagents.com/setup.sh -o setup.sh
less setup.sh
bash setup.shYour keys stay yours. The script runs locally and talks only to the official sources it names (nodejs, Ollama, GitHub). Any key you enter is typed invisibly and written to one file on your machine —
.env, readable only by you. It is never uploaded, logged, or shared, and nothing installs without your explicit yes. The one honest caveat: Ollama's own Linux installer uses sudo internally to register its service, and the script says so at that step before asking.
What the walk-through covers
Six steps, each skipped automatically if your machine already has the piece, and each optional — answering no to everything changes nothing. The whole run takes about ten minutes, most of it the model download.
The pieces, and why the course wants them: git fetches the public course repo. Node.js 22+ runs the agent framework. Ollama runs open-weight models on your own machine, which is how the course's opening hands-on modules run with no account and no API key. qwen3:8b is the course's workhorse model, a 5.2 GB one-time download that is billed never. The repo (github.com/jhwadman/melchizedek-agents) holds every agent the modules study. And .env is the one local file your keys live in; the free Gemini key that Part 2's later modules use can be added there now or weeks from now.
What this course is — and what it is not
This course teaches you to build
- real agents, defined in readable configuration, running on models you own or rent by choice
- systems you can deploy inside any application you're building — the patterns stand on their own in any stack
- the full production discipline: instruction anatomy, testing, memory, tool reach, visual review, agentic coding
- judgment — how to design, verify, and trust an agent's work, which no vendor UI can do for you
Taught well elsewhere
- how to use Claude Code, Cursor, or any particular coding assistant
- how to prompt ChatGPT better — chat technique is a different craft
- vendor certifications or tool-of-the-week tours
- software engineering itself — you need a terminal and a config file, and the course carries you from there
The distinction matters because tools age and mechanisms don't. A course on today's chat window expires with its interface; the agent loop, the grounding doctrine, the critic gate, and the memory record survive every rebrand. You leave able to build the thing, not just operate someone else's.
Other roads in
By hand: every module states exactly what its specimen needs, and the curriculum page lists the full requirements — Ollama and the model for the open-weight modules, the free Gemini key from module 2.04. The script automates precisely those steps and nothing more.
With your coding agent: if you already run Claude Code or a similar tool, hand it melchizedek-agent-setup.md — a paste-ready prompt that has the agent do the machine steps while the human steps (accounts, keys) stay explicitly yours.
On Windows: the script runs in WSL, which is Microsoft's supported Linux environment and takes one command to enable. Install WSL, open its terminal, and run the same curl command there.
And if you install nothing at all, the course still teaches: every hands-on module carries scripted traces that reproduce the real model's behavior faithfully, so you can read first and build the day you're ready. The door is on the syllabus.