Build the agent yourself.
The Lyceum's course and a studio for agents, in one desktop app
Coming soon
Melchizedek for the desktop
The Lyceum's course is moving into a desktop app. It will hold the whole course: every module with its checkpoints, the questions that ask for your own answer, the replayable traces and the instruments. Beside it is a studio where you open the agent teams the course studies, run them and read every model and tool call they made.
The app will be free. The models are not: you bring your own keys, for whichever providers you choose, and they stay encrypted on your machine. A module that runs on open weights needs no key at all. Until it ships, the whole course is here to read.
Academy
The course with everything the reading edition leaves out: the checks, your saved answers, the traces to replay and the hands-on runs, each beside the module that teaches it.
Studio
Open an agent team from its definition file, chat with it, and see the orchestration as it runs: which agent answered, which tools it called, what each call cost.
Or read the course here.
The reading edition is the same course as prose and figures: how a language model computes its next word, how to design, test and orchestrate agents, and how to grade them. It asks nothing of your machine.
Understand what a model computes, run open weights on your own hardware, and author the system prompt that commits it to one operational purpose.
10 modules · ~15–75 min each Agent Design and EvaluationDefine whole agent teams in configuration files, with persistent memory, live tool execution over MCP, and regression testing before deployment.
13 modules · ~18–47 min each Mastering the Principles of LLMsTransformer mechanics at the level of arithmetic: what the weights encode, how context conditions the distribution, why fabrication is the default, and where a fact belongs.
11 modules · ~16–52 min each