E1 · Design your Jarvis
15 minWrite the meu-jarvis.md: name, style, what it should do, what it should never do, and what it needs to remember.
Done when: the file fits on one screen, and someone else could understand who your assistant is.
Lesson 3 · 10 min
Step 2: your Jarvis with INTELECTO
A personal assistant written from scratch in Python, kept small on purpose: it talks with you on Telegram, remembers your conversations, and uses a cloud or local model. You stop opening a website to chat: the assistant is yours.
Start lesson
In this lesson, you will
- 1Design the assistant before installing it
- 2Create the bot on Telegram
- 3Test the memory
- 4Choosing between cloud and local
Objective: design, install, and put to work an assistant with memory and a personality you define.
01Your assistant’s architecture
From Telegram to the model and back.
Scroll down: each step below lights up its part of the diagram.
- 1/5
You
You chat on Telegram, on your phone or computer, like you would with any contact.
- 2/5
Telegram bot
The bot is the entry point. You create it with @BotFather using /newbot and store the token only in .env, as you learned in step 1.
- 3/5
INTELECTO
INTELECTO receives the message, combines your assistant’s personality with what it remembers about you, and puts together the request for the model. It’s small by design, so you can understand and change it.
- 4/5
Model
The model generates the response. In the cloud (OpenRouter), it’s more powerful and costs money per use; locally (Ollama), it runs on your computer, if it can handle it.
- 5/5
The full conversation
You write in Telegram, INTELECTO combines personality and memory, the model responds, and the response returns to the same chat. It’s all yours, with the token stored in .env.
Want to see other assistant systems ready to install? The Events JARVIS section brings several together.
02From the design to your routine
Click each step.
1. Design
Before installing, write down who the assistant is: its name, how it speaks, three things it should do every week, three things it should never do, and what it needs to remember about you.
This file answers the installation questions.
Test
Someone else reads the file and understands who your assistant is.
2. Create the bot
On Telegram, open @BotFather, send /newbot, and choose a name and @username.
Store the token only in .env.
Step 1 rule
The token doesn’t appear in any file besides .env.
3. Install
Clone INTELECTO, export your briefing from the comparison page, and ask Claude Code or Codex to build from it and doc/intelecto.md, with tests.
Use the answers in your meu-jarvis.md. The full step-by-step guide is in lesson 4 of the step 2 course.
Done when
You send "hi" and it replies with the name and style you chose.
4. Memory and routine
Tell it three facts about yourself one day, then ask about them the next day in a new conversation.
Then turn one weekly task into a standing request and use it five days in a row.
Standing request
Every morning, when I send "day," list my three priorities based on what I told you yesterday.
03The setup template and installation
Copy the template, fill it in, and only then install.
meu-jarvis.md
# My Jarvis ## Name and speaking style Name: How it speaks (formal, direct, humorous): ## What it should do every week 1. 2. 3. ## What it should never do 1. 2. 3. ## What it needs to remember about me - Profession: - Routine: - Preferences:
Terminal · download and run the example
git clone https://github.com/inematds/intelecto.git cd intelecto/exemplo-jarvis python3 -m unittest discover -s . -p 'test_*.py' -v python3 smoke.py
Repository example
- The Telegram token (from @BotFather), only in .env.
- The model: cloud-based (OpenRouter) or local (Ollama).
- The personality: your meu-jarvis.md.
Run the same question with both and record the results in a table:
- Response time
- Response quality
- Cost per use
The goal is to use your own numbers to know which model to use for what.
04Practice: the 6 exercises in sequence 2
Prerequisite: complete sequence 1 (terminal, folders, .env).
E2 · The Telegram bot
10 minCreate the bot with @BotFather using /newbot and save the token in .env.
Done when: you have the token, and it doesn’t appear in any file except .env.
E3 · Install
30 minClone INTELECTO, export the briefing, and ask the agent to build your Jarvis with tests (lesson 4 of the step 2 course).
Done when: you send "hi" and it replies with the name and style you chose.
E4 · Memory test
2 daysDay 1: tell it three facts about yourself. Day 2, in a new conversation: ask about them.
Done when: it remembers without you repeating yourself.
E5 · Cloud or local
20 minAsk the same question with a cloud-based model (OpenRouter) and a local one (Ollama). Record the time, quality, and cost.
Done when: you can say which model to use for what, based on your own numbers.
E6 · A real routine
20 minTurn one of the tasks from E1 into a standing request and use it for a week.
Done when: you used it 5 days in a row and noted what to improve.
Full exercises: jornada-agentes/exercicios/02-jarvis.md. The extra challenge (the task list you send before going to bed) is the start of step 3.
05Sequence 2 on video
INTELECTO on Telegram, with memory and personality.
Create your Jarvis · 1 min 33 secVideo page with script →
06Test what you learned
The explanation appears after your answer.
Question 1 of 3
Why write the meu-jarvis.md before installing?competency
Correct
Design comes first: with its name, style, what to do, and what never to do written down, you can answer the guided setup without improvising and have a way to check whether the assistant is right.Not quite
Design comes first: with its name, style, what to do, and what never to do written down, you can answer the guided setup without improvising and have a way to check whether the assistant is right.Review: From the design to your routineQuestion 2 of 3
The assistant didn’t remember what you told it yesterday. What’s the first step?
Correct
Exercise E4 says exactly this: if it doesn’t remember, check the INTELECTO course to see how memory is stored and adjust it.Not quite
Exercise E4 says exactly this: if it doesn’t remember, check the INTELECTO course to see how memory is stored and adjust it.Review: From the design to your routineQuestion 3 of 3
When does a local model (Ollama) make sense?
Correct
Make the choice based on measurements: ask both the same question, make a table with speed, quality, and cost, then decide based on your numbers.Not quite
Make the choice based on measurements: ask both the same question, make a table with speed, quality, and cost, then decide based on your numbers.Review: The setup template and installationTest complete
You got {a} out of {n} right. Get them all right to mark the lesson as complete.Choose an answer. The explanation appears after your attempt.