The labs
Take it apart and see
Ten instruments, one system. Each one isolates a single part of how an AI application works so you can change it and watch what happens. They are ordered as a path, but any of them stands alone.
- Lab 01Token VisualiserType anything and watch it split into the units a model actually reads.Open
- Lab 02Context Window LabFill a context window and watch what gets pushed out when it overflows.Open
- Lab 03LLM AnatomyStep through the conceptual journey from text to the next token.Open
- Lab 04Prompt PlaygroundBuild a prompt from its parts and see what each part changes.Open
- Lab 05Agent LoopStep an agent through observe, decide, act, and observe again.Open
- Lab 06Tool Calling LabAsk for something a model cannot do alone, and watch a tool get called.Open
- Lab 07RAG LabAsk a question and watch documents become retrieved context.Open
- Lab 08API ExplorerBuild a request, see the response, and read what every field means.Open
- Lab 09MCP & ConnectorsFollow a request from an AI app out to an external system and back.Open
- Lab 10Agent LabAssemble an agent from its parts, run it, and watch every step.Open
- Lab 11Workflow LabRun one real-shaped task through every stage you have just learned.Open
What is real and what is not
Every lab states plainly what it is doing. Tokenisation follows real tokeniser rules but is not any specific model's vocabulary. Retrieval uses keyword overlap rather than embeddings, and shows you where that fails. Tool results come from fixed sample data, except the calculator, which genuinely computes. Unless an AI provider is connected, no language model is called anywhere on this site — and where output is generated by rules rather than a model, the label says so.