Study notes
AI education
I wrote these to get better at local models and the systems around them. The path is the one I study: build the mechanism, measure it, and only then pick a library. The same notes are in English and Polish.
Lessons CC BY 4.0 · code MIT
Start the second course after part 6 of the first. The agent loop is what the rest stands on.
Courses
10 parts · ~47 h
From a model to your own agent
Tokens, sampling, memory, senses, then the agent loop. You build each layer before you trust a library.
Start · Course: from a model to your own agent — index19 parts · ~102 h
Agentic applications
One system you can leave running. Memory, tools, a sandbox, a schedule, evaluation, and a hard line on what it may do.
Start · Course 2: agentic applications on local models — indexSurveys
20 harnesses
Agent harnesses
What already exists. Twenty open harnesses that turn a local model into an agent, and when you should still build your own.
Start · Agent harnesses for local models — an overview15 assistants
AI assistants
Fifteen assistants, how they are built, how they charge, and what that means for a local assistant in Poland.
Start · AI Assistants — the Top 15 and What to Learn from Them20 routes
Cheap and free models
Where online inference is cheap enough for an agent, and what still has to stay on your own machine.
Start · Cheap and free online models for agents — the top 20 routes (Oct 2026)