Learn AI by building one, not by reading about it.
Adventure I gets you from zero to a real running AI agent in 8 hands-on lessons. Adventure II opens the LLM black box — tokens, attention, prompts, agents, MCP, and shipping a real product. Workshops are one-sitting deep-dives. Sandboxes are runnable mini-sites you can inspect and tinker with.
01 · Adventures
Pick your path
Each adventure is a multi-lesson arc — 8 lessons in Agent, 6 in LLM. Start with Agent if you want to ship something fast — start with LLM if you want to know how it really works.
Build Your AI Agent
Eight hands-on lessons that take you from zero to a real AI assistant — running on your own server, talking to your phone, never forgetting a thing.
Inside the Large Language Model
A 6-week course on what's actually inside the machine that talks — tokens, attention, transformers, and how words become math.
02 · Workshops
Standalone workshops
One topic, one sitting. Pick a workshop when you want a focused deep-dive without committing to a full Adventure.
03 · Sandboxes
Hands-on sandboxes
Just want to play? These are tiny working sites you can open, inspect, and modify. Each comes in English and Chinese.
My First Website
A self-contained HTML page: 3 static cards + 1 fetch-powered card. Built to be inspected and remixed.
Open sandbox →Next-Token Predictor
Watch an LLM-shaped probability distribution shift in real time as you change the temperature. Click to sample one token.
Open sandbox →Tokenizer Playground
Type any sentence and see how an LLM splits it into tokens. Compare English vs Chinese costs side-by-side.
Open sandbox →Prompt Pattern Lab
Toggle 5 prompt patterns and watch the same AI give a far better answer. The model didn't get smarter — your prompt got clearer.
Open sandbox →Agent Loop Simulator
Step through the think → call a tool → read the result loop. The model can't do math or check the weather — but it knows which tool to call.
Open sandbox →MCP Plugboard
Plug MCP servers into an agent like USB devices and watch its tools appear. Write a tool once, every agent can use it.
Open sandbox →PRD Forge
Turn a product idea into a one-page PRD plus a 'ship the empty URL' checklist. A product is one core loop other people can reach at a URL.
Open sandbox →Lie Detector
The AI sounds 100% sure — but is it right? Catch it confidently making things up, and learn why a high confidence bar isn't truth.
Open sandbox →Meaning Map
AI turns words into points in space — close together means similar meaning. Tap a word to light up its neighbours, then watch king − man + woman ≈ queen.
Open sandbox →Token Meter
AI charges by the token — and every output token costs more than an input one. Pick a prompt and watch the bill add up across model tiers.
Open sandbox →Attention Spotlight
Click a word and see which earlier words it points back to — that's attention. Change one word and the spotlight visibly jumps.
Open sandbox →Library Detective
Give the AI the right pages and it stops guessing — and cites its sources. Hide the library and watch it confidently get it wrong. That's RAG.
Open sandbox →Noise to Picture
Image AI doesn't paint left-to-right — it starts from pure static and cleans it a little each step, guided by your words, until a picture appears.
Open sandbox →Train a Tiny Brain
Watch a model learn live: nudging just three numbers until one line splits the two teams of dots and the loss drops. That's training, at the smallest scale.
Open sandbox →Cloud Command Deck
SSH into a pretend cloud server and explore with real Linux commands — ls, cd, cat. Learn what a server is, and why your AI agent never sleeps.
Open sandbox →Soul & Memory Lab
Edit an AI's SOUL.md and USER.md, then restart it to watch its personality and memory load from plain text files. No file, no memory.
Open sandbox →API Explorer
Order from a pretend API menu and watch JSON come back. Learn that an API is a menu, a GET request just asks, and JSON is data in a tidy, labeled format.
Open sandbox →Message Journey
Send a chat message and walk it across the internet — your phone (client) → Telegram → your cloud server (OpenClaw) → the AI → back home. See why the API token is your bot's password.
Open sandbox →Cron Scheduler
Build a cron expression from five tiny fields — minute, hour, day, month, weekday — and watch the next fire times appear. Learn that a cron job is a task plus a trigger, and that your always-on server runs it on time while you sleep.
Open sandbox →Ask Your Data
Meet a knowledge base — just a table of rows and columns. Tap a plain-English question and watch the AI turn it into SQL, run it on the `pets` table, and light up exactly where the answer came from. Your agent looks things up instead of guessing.
Open sandbox →Tool Router
MCP hands the AI a toolbox — 🧮 Calculator, 🌤️ Weather, 🕐 Clock, 📖 Dictionary. Ask a question and watch the AI pick the right tool itself, then chain two together to compare two cities. Unplug a tool and it can only guess.
Open sandbox →Agent Assembler
The capstone: a real AI agent isn't one magic thing — it's the 7 pieces you built across Lessons 1–7. Plug them all in, then watch a real morning briefing flow through every piece. Pull one out and see exactly where it breaks.
Open sandbox →