Academy
Become the maker
You don't need to be technical to build real things any more. You need a clear idea and the knack of working with an AI agent. The Academy teaches that knack — short, practical lessons that each end with something actually working.
Describe what you want on one side; a working system — a website, an automation — comes out the other. No middleman.
Our job here isn't to turn you into a programmer. It's to show you how to drive the agent, where to point it, and how to fix things when they sulk — so your ideas become real without waiting on anyone.
The what and the how
Here's the shape of the whole craft. Modern AI does something remarkable: you tell the computer, in plain language, what you need — and it does it. How to tell it you'll pick up here in minutes, hours at most. The bigger half is what to tell it: which pieces a site needs, how they behave together, how it connects to other systems so it stays reliable and secure. We teach both — the how and the what of working with AI.
Organic, not by the textbook
Most courses teach the parts one at a time — a lesson on databases, a lesson on logins, a lesson on payments. We don't. We start from a project you actually want to exist, and add whatever it needs, when it needs it. If your site suddenly needs a database, we bolt one on and keep going — no ceremony, no separate course about it. You won't use every part to the full, and that's fine: it's there for when you do. You learn the pieces by building something real, in the order your own project asks for them — which is exactly how it works once we're out of the picture.
Fresh, and honest about it
The AI world moves fast — faster than any lesson can. Something we taught here in good faith, even something we walked through with students last month, might already work a little differently today. We keep the lessons current, but the ground genuinely shifts under all of us. So if you hit a step that doesn't match what you're seeing on your screen, that's not you getting it wrong — it's the world moving. Tell us and we'll fix it fast. A quick note via the contact form is plenty — we'd genuinely rather hear it from you.
Why we do this — and what's always free →
One student free each month — we teach someone free →
What's always free, and what's not →
Glossary — plain-language tech terms →
Index (rejstřík) — every term, everywhere it appears →
The path — seven levels, then a project track
The Academy is a ladder. Each level moves you from "I can do X" to "my agents do X for me." Every level starts with one tiny first win — a little "zero project" you build in minutes — then you remix it into your own thing and climb on. Start at Level 1 and go in order; you'll feel each rung click.
Level 1 · Build your website Free
Say what you want — minutes later it's live and looks great. The whole stack, from an empty computer to a public, design-rich site.
1.1 · From zero to a live website
Od prázdného počítače k vlastní stránce na internetu — klidně do hodiny. Účty, nástroje, první web napsaný agentem, živý na Cloudflare. Interaktivní, pro úplné začátečníky. (CZ)
1.2 · Working with an AI agent
The simple loop that builds almost anything — describe, build, check, feed back, refine — and the one debugging habit that makes it click.
1.3 · Models, limits & cost
Which AI model to reach for and when — fast and cheap for routine, strong for the hard parts — and how to stop burning your daily limit before you learn.
1.4 · Edit & publish
Change something, push it, and watch Cloudflare put it live — the git-as-deploy-trigger loop that makes the site truly yours to run.
1.5 · Ship something real: a form
Put a working, spam-protected contact form on your site using our step-by-step builder. Your first proper feature shipped.
1.6 · Reading errors without fear
What error messages really are, and how to turn any of them into a fix in two minutes — the habit that makes everything above stick.
Level 2 · Minimize the clicking Free
Turn Level-1 steps into things that happen by themselves. Teach the agent a skill once and it re-runs it forever — kept in its own git repo. And learn why every page carries a build stamp.
2.1 · What is a skill
How an agent learns a repeated job once and re-runs it on its own — and why skills live in their own git repo so they travel with you.
2.2 · Keyboard shortcuts for your site
Make your site feel grown-up: ⌘K / Ctrl+K to search, "/" to search, Esc to close, ? for help — done accessibly. The exact pattern we use here, ready to ask your agent for.
2.3 · The build stamp, explained
Release date + UTC time + build version on every page — what it's for, and why traceability saves you when something looks out of date.
ComingLevel 3 · Projects multiply Free
A bit boring, but tidy before you scale. One standard way to organize every project — readme, claude.md, a wishlist, a changelog — with automatic git backup and encrypted keys so you can move to a new machine after a crash.
3.1 · The standard project layout
readme · claude.md · wishlist · changelog, how those rules inherit down folders, hidden files demystified, plus automatic git backup and encrypted-key recovery.
ComingThe line · above is free, below is Members'
Everything you need to build is free. The rest is letting go.
The whole craft of making — building real things, shipping them, running them yourself — lives above this line, Levels 1–3, and it's yours at no cost, forever. That's not a teaser; it's genuinely enough to ship.
Below the line is the Members' layer — the next gear: handing the work to your agents and walking away while they get on with it. The price isn't a goldmine and it's no Klondike — it's a contribution to keeping the system running, the lights on and the lessons current. If that layer is for you, you're warmly welcome.
Join the enlightened — see what's free vs. Members' →No pressure: if you're just here to build for yourself, stop at the line. The free craft above is the whole craft of making.
Level 4 · Autonomous agents Members'
Stop being the agents' postman. Marschall watches them so they work when they're idle and have things to do; an agent-bus lets them talk; a kanban board they drive themselves. They get on with it while you're away.
4.1 · Agents at full autonomy
Stop being your agent's bottleneck. An autonomy contract — act on the reversible, gate only the irreversible — inherited across every agent. The productivity unlock most people never reach.
Members'4.2 · AI-native craft
The seven best-practices that make an agent dependable instead of flaky — evals, observability, dry-run defaults, guardrails, least privilege, decision logs, versioned prompts. We adopt them, and we teach them.
Members'4.3 · The escalation reflex
What an agent should do the moment it's stuck: look for a skill, and if none exists, ask for one to be built — instead of flailing or dumping the problem back on you. The habit that turns dead ends into tools everyone reuses.
Members'4.4 · The autonomous engine
A scheduled "drain my lane" task that runs on its own clock — no chat open, no one watching. The drain loop, the impact boundary (what ships alone vs waits for you), and keeping the machine awake.
Members'4.5 · Solo, Duet & Double
The three ways a chat can answer — one agent, two different ones, or two identical ones — and how to pick on purpose: Solo for cost, Duet for quality, Double for stability. Stop paying for reliability you don't need.
Members'4.6 · Bus, board & Marschall
Wire the agent-bus, the self-driven kanban board and the Marschall watcher + scheduler — the war-stories included, so your team works while you sleep.
Members' · comingLevel 5 · Reporting in your pocket Members'
A password-protected admin in your private web: AI spend in Kč and the board, checkable from a tablet or phone — no longer tied to the machine running cowork.
5.1 · Your admin cockpit
A phone-checkable, password-protected admin: AI costs in Kč plus the board, on the non-public side of your site.
Members' · comingLevel 6 · Orchestrate tools with n8n Members'
Let agents build whole flows: wire nodes, set credentials, have flows call each other. The leap from "one task" to "a system of tasks running on their own."
6.1 · Your first automation
The shape of every flow — trigger → AI brain → action — built around a real one: new enquiry, triaged by AI, sent to the right person.
Members'6.2 · Agents that build flows
How to point an agent at n8n and have it wire whole flows on its own — credentials set, nodes connected, flows calling each other.
Members' · comingLevel 7 · When the sandbox can't reach Members'
Where a sandboxed agent can't reach a service directly, route the work through n8n — using credentials from your orchestration tools, made available to agents safely, for the complex missions.
7.1 · Send n8n with credentials
Routing "the sandbox can't reach this" work through n8n, and making orchestration credentials available to agents — without handing over the keys to everything.
Members' · comingLevel 8 · Build your own Health OS — the idea Free
Your measurements, your database, your agents — and a place to learn Pages, Functions, D1 and MCP where being wrong costs nothing. The architecture, the reasoning, and the four bugs that cost the most.
8.1 · Why your own data is the best place to learn
Almost zero stakes, messy multi-source data, a real privacy boundary — and skills that transfer straight to invoicing, support and bookkeeping.
Free8.2 · The shape of it — architecture without code
Static pages, Functions, D1, MCP, two auth paths — and a public tier that redacts on the server instead of hiding in the UI.
Free8.3 · When failure looks like success
Four real bugs that all reported success: partial imports, CSS masking a bug, an exclude-list failing open, and timestamps that were never uniform.
FreeLevel 9 · Build your own Health OS — the build Members'
The week of trial and error, skipped: schema with idempotent imports, Functions with a default-deny middleware, MCP tools an assistant can actually use, scoped OAuth, and one composite number that admits what it does not know.
9.1 · The schema — one table per domain
Why not one events table, why day is stored twice, and the partial unique index that makes a repeated import a no-op.
9.2 · Functions as an API — and the public/private split
File-based routes, one default-deny middleware, and redaction that happens in the SELECT rather than in the browser.
Members'9.3 · Tools, not copy-pasted JSON
An MCP server of narrow, purpose-built tools — validated at the boundary, logged on every call, and impossible to misuse into a bad write.
Members'9.4 · Two doors, one room — OAuth and bearer tokens
A scoped, revocable grant for a third-party assistant; a static token for your own agents; one check for both.
Members'9.5 · One honest number
Weighted pillars, ramps instead of thresholds, weight redistributed when data is missing — and why a score must report its own coverage.
Members'9.6 · Importing real devices without corrupting yesterday
Raw first, aggregate never replace, idempotency keys, five devices disagreeing about what a day is — and a pipeline that shouts when it did nothing.
Members'9.7 · Where the course ends
What you built, the same pattern applied at work, the parts deliberately left out, and when a consultation beats another weekend.
Members'