00. About the Course

The mini-course “Agent from Scratch” – we learn from the book’s notes and immediately assemble our mini-agent: a console program that decides on its own when to call a tool, calls it, looks at the result, and repeats until the task is solved. This is approximately how ZCode orchestrates: task → reasoning → tool call → observation → again.

What will be the outcome

A working CLI agent (~60 lines): you enter a task in the terminal – the agent calculates, finds out the time, and responds with text. The code is in Python and JavaScript (Node), choose your language or do both.

How the course is structured

Each lesson is one idea: a short summary, a block diagram, links to atomic notes from the book (theory – it’s there, read it in full) and a short code example in Python and JS. Practice builds up: by lesson 5, your agent is assembled from pieces of lessons 1–4.

Course Roadmap

1. [Lesson 1. Agent Formula] – model + context + tools
2. [Lesson 2. ReAct Cycle] – reasoning → action → observation
3. [Lesson 3. Harness and Guardrails] – three protective layers
4. [Lesson 4. Tools] – how LLM calls functions
5. [Lesson 5. Assembling the CLI Agent] – all the code in one piece and running it

What you’ll need

• Python 3.10+ or Node.js 18+ (no additional installation required in both cases)
• A key for any OpenAI-compatible API: OpenAI, Groq, POLZA, local Ollama – anything works
• Ability to read simple code; writing is not necessary, but better

Questions – in the comments of the Telegram channel (✈ in the sidebar).

Related: [Learning], [Agent = LLM + Context + Tools]