E3a · Run the example
10 minClone INTELECTO, go into exemplo-jarvis, and run the tests and smoke.py.
Done when: All 9 tests pass, and the smoke test shows the DB line with the name.
Lesson 4 · 20 min
You don't write code. You choose the ingredients on a page, like at a grocery store, export a list, and ask Claude Code or Codex to build it by following the INTELECTO construction plan. Then you check it with tests and a conversation in the terminal, before connecting it to Telegram.
Start lesson
Goal: generate the brief at the grocery store, ask the coding agent to build it with tests, and check the core with tests and a conversation.
Click each block. The agent does the heavy lifting.
Start small: a few ingredients, passing tests, then grow. You can go back to the grocery store and add more.
Click each step.
For a first Jarvis, the installation video chose eight: personality file, Telegram bot, WhatsApp, memory with SQLite search, encrypted secrets vault, list of forbidden commands, scheduling, and skills.
You can start with less: personality, Telegram, memory, and vault.
Your cart has 4 to 8 items, and you can explain what each one does.
Export the brief for Claude Code and download meu-assistente.md. Clone the INTELECTO repository and put the brief inside the folder.
What the tests revealed: the list alone isn’t enough. What makes the build work is the pointer to doc/intelecto.md.
intelecto/ with doc/intelecto.md, reference-repos/, and your meu-assistente.md.
First ask for the plan: the agent reads everything and returns the files, order, and questions (Docker or direct install? OpenRouter or Ollama? Which model?). Reply in Portuguese.
Once you approve the plan, ask it to build with tests. Approve each command it wants to run only when you understand what it does.
Describe the problem in Portuguese and ask it to fix it. This cycle is normal and expected.
Run the tests. Then chat in the terminal: "remember my name is ...", followed by "what's my name?".
The last line of the example shows the name saved in the database: proof that it isn't fixed text, and that memory and the loop work together.
All tests pass, and the name appears on the DB line.
Run these in the terminal, from your projects folder.
git clone https://github.com/inematds/intelecto.git cd intelecto # copy here the meu-assistente.md that the grocery store downloaded
Read meu-assistente.md and the plan in doc/intelecto.md. Before writing code, give me the plan: files, order, and decisions that depend on me. Use my identity (SOUL, AGENTS, USER, and MEMORY in workspace/).
Plan approved. Build my assistant in Python, with tests, following the contracts in doc/intelecto.md. The Telegram bot replies only to my user. Never put secrets in the code. When you're done, run the tests and show me the results.
cd exemplo-jarvis python3 -m unittest discover -s . -p 'test_*.py' -v python3 smoke.py
USER : remember my name is Nei JARVIS: Your name is Nei. USER : what's my name? JARVIS: Your name is Nei. DB : ['Meu nome é Nei']
The DB line proves that the name was saved in SQLite; it didn't come from fixed text.
Exercise E3 from sequence 2, following the path the repository actually uses.
Clone INTELECTO, go into exemplo-jarvis, and run the tests and smoke.py.
Done when: All 9 tests pass, and the smoke test shows the DB line with the name.
At the grocery store, select 4 to 8 ingredients and export meu-assistente.md to the INTELECTO folder.
Done when: The file is in the folder, and you noted the reason for each ingredient in your meu-jarvis.md.
Copy your workspace/ folder from lesson 2 into the project, make requests 1 and 2, and follow along with the agent.
Done when: The tests pass, and in the terminal conversation, it replies with the name and style you defined.
In the original exercise, the step is ./start.sh. In today's repository, the path is: brief + plan + agent builds. The setup assistant is one of the pieces the plan says to build.
From the grocery store to an assistant that replies, with the whole process on screen.
The explanation appears after your answer.
Question 1 of 3
Why give the agent meu-assistente.md along with doc/intelecto.md?competency
Question 2 of 3
The agent finished and said "done." What's the proof it works?
Question 3 of 3
What’s the right way to get started?
Choose an option. The explanation appears after your attempt.