lyceum@agents

Build the agent yourself.

Interactive lessons for the age of autonomy

The course

18 modules · 3 parts · ~12 hours · no account — every exercise runs in your browser

Start the course ▸

LLM Fundamentals

You learn what a model actually computes, run one on hardware you own, and then open it up — down to the vector that chooses each word.

8 modules · ~30–40 min each Agent Design

One written instruction commits a model to your purpose; one file defines a whole team — with memory, live data, and code it can be trusted to write.

9 modules · ~35–45 min each Agent Observability

Grade a change over real cases, calibrate the judges that grade them, record every turn, and let the deploy step refuse a regression.

1 module · ~50 min each

Tooling

We build open-source instruments alongside the writing, starting with melchizedek, a multi-model, multi-agent orchestration framework — and almost every module of the course studies a living specimen from it.

  • The framework is open source; clone it and every lesson's team runs on your machine
  • Each module hands you its specimen's verbatim agent definition as a download
  • The hands-on track opens on open weights — no account, no meter, no data leaving your desk

→ all tools

the Tutor's first turn — a live local run, compressed

$ npm run syndicate:tutor

Syndicate: Tutor · Orchestrator: Tutor (ollama/qwen3:8b)

you ▸ The stoa was a covered walkway in Greek architecture… — Teach me the key idea in this material.

Tutor ▸ The material connects two things: the physical structure of the Stoa (a covered walkway) and the philosophical school of Stoicism.

What do you think the significance of the Stoa Poikile's location might have been in Zeno's teachings?

module 1.02 replays the full turn, law by law