friendly lessons, working code, and room to experiment
Whether this is your first model call or your fifth agent, you’re welcome here. Start with the syllabus, choose the lesson that sounds useful, or simply begin at lesson one. Everything stays in the archive, so there’s no need to rush.
Give the lesson 3 loop a pair of eyes. We’ll send it a screenshot, help it tidy a layout, and use a simple DOM check to confirm the page is ready. JavaScript and Python examples run on Playwright.
Help your agent join a wider team. We’ll publish an agent card and handler on Blocks.ai, ask another agent for help, and bring the answers safely back together. JavaScript and Python examples share one 60-second budget.
Give your agent a clear finish line. We’ll set a turn limit, write a goal your program can check, recover from slow tools, and verify the result before the loop says it is done.
Give a chat model its first useful tool. We’ll build the loop by hand in JavaScript or Python, follow one complete trip around it, and see how a friendly, precise description guides the model’s choices.
Begin with a real language model running on your own GPU—no account, key, or server needed. We’ll inspect what goes in and comes back, build a one-file demo, and gently explore the limits of browser inference.
Meet the whole course before you begin. You’ll see what each lesson builds, what you need, and how WebLLM, tools, loops, communication, vision, and evals fit together.
A warm invitation to learn at a comfortable pace: one idea, working code, and a browser exercise each week. The course is free, needs no account, and works with local models.
Come meet the little machine behind the lessons: one open-source page with a virtual filesystem, undo, previews, and up to 23 tools. It runs on desktop or phone, and it is yours to inspect and change.