Finish the check
5Run verify after configuring. If an item is missing, resolve the indicated issue and run it again.
Done when: The last line says “everything looks good”.
Lesson 2 · 16 min
Install and chat in the terminal
You already downloaded the project in lesson 1. Now choose a local AI that fits your computer, answer the seven setup questions, and check the installation before chatting.
Start the lesson
In this lesson, you’ll
- 1Prepare Python and the local model
- 2Choose a brain and approval policy
- 3Check the installation
Objective: Set up INTELECTO with Ollama locally and confirm that verify finishes with “tudo certo” (all good).
01From command to response
Four parts are enough for the first conversation.
Click each block in the diagram to see what it does.
Click each part to see what it does.
02First use, in order
Click each step to see what to do.
1. Go to the project
Go to ~/intelecto-v2. If you haven’t downloaded the project yet, clone it as shown in lesson 1.
2. Install the brain
On Linux, install Ollama with the official command below; on macOS, download the app from ollama.com. Then use ollama pull to download your chosen model.
Live test on 11/13/2026: with llama3.2, “remember that my name is Nei and I live in Porto Alegre” was saved directly by the assistant and recalled in a new conversation. This works even when the small model doesn’t call the memory tool. It still makes little use of other tools, such as files and commands. qwen3.6:35b-a3b called tools in the test, but its local file is much larger.
3. Choose options
The seven answers: name and personality in your own words; brain: ollama; model: one of those listed by configurar; where to run commands: local; approval: perigosos; Telegram IDs: leave blank for now. Press Enter to keep the value in brackets. Telegram is a conversation channel, not a model provider.
The executor is the program that runs the command; choose local for now and see the Docker box in lesson 6. OpenRouter is a paid cloud service and isn’t part of this exercise.
The approval policy has three choices: perigosos (dangerous; the default) lets known read actions run and asks about other actions; sempre (always) asks before any command, including ls and cat; nunca (never) lets known read actions run but refuses anything that would require approval. Commands blocked by the safety lock won’t run under any option.
4. Check and chat
verify checks what your current configuration requires. “Everything looks good” doesn’t prove the model will use tools; have a real conversation to check. SQLite with FTS5 means local memory with text search. Fernet is the encryption mode for the vault when the cryptography library is installed. On Ubuntu, if you want Fernet for the vault, install python3-cryptography with the package manager; pip install cryptography may fail with externally-managed-environment on Ubuntu 24.04.
03Which model fits your machine?
Combined sizes of the files listed in the installed model records on this machine on 11/13/2026; the download size may vary. These are not RAM measurements.
| Your machine | Model | Measured local file |
|---|---|---|
| Everyday computer | llama3.2 | 2.02 GB (2,019,392,628 bytes). Remembers facts when you say “remember that…”; makes little use of other tools. |
| High-end graphics card | qwen3.6:35b-a3b | 23.94 GB (23,938,333,115 bytes). Called tools in the live test. |
Start with a model your machine can download and run. The file size doesn’t guarantee it will fit in memory. llama3.2 remembers facts you ask it to remember with “remember that…”, even on an everyday computer; to use other tools, check whether the model actually called them.
04Project commands
Copy the commands and run them in the terminal from the specified folder.
Terminal · go to the lesson 1 folder
cd ~/intelecto-v2
Ubuntu · Fernet for the vault (optional)
sudo apt install -y python3-cryptography
Linux · install Ollama
curl -fsSL https://ollama.com/install.sh | sh
Terminal · download the everyday model
ollama pull llama3.2
Optional · model for a high-end graphics card
ollama pull qwen3.6:35b-a3b
Terminal · configure and check
python3 -m intelecto configurar python3 -m intelecto verificar
Terminal · chat
python3 -m intelecto
Repository example
The test used the mock provider (simulated brain) and a temporary folder. Only the path was normalized to show where it will be on your computer: ~/.intelecto2. This does not verify Ollama.
[OK] Python 3.12.3 (precisa 3.10+) [OK] SQLite com FTS5 (memória) [OK] config em ~/.intelecto2/config.json [INFO] cofre: fernet (Fernet) tudo certo
The program prints this in Portuguese, exactly as shown.
05Your first conversation
Follow the steps in order and check each one only after you’ve confirmed the result.
Talk to the assistant
5Open the chat and ask a simple question. The model must be available in Ollama.
Done when: The assistant replies in the terminal, and you can exit by typing “sair”.
06Check your understanding
Choose before opening the explanation. If you get it wrong, use the link to review.
Question 1 of 3
What does the verify command check?
Correct
It checks the current configuration. It doesn’t replace testing a real conversation or verify Telegram live.That’s not it
It checks the current configuration. It doesn’t replace testing a real conversation or verify Telegram live.Review: First use, in orderQuestion 2 of 3
Which local provider was tested live with memory, including llama3.2?
Correct
Ollama is the local provider used in the live test. OpenRouter uses the cloud; mock simulates responses and isn’t a real local model.That’s not it
Ollama is the local provider used in the live test. OpenRouter uses the cloud; mock simulates responses and isn’t a real local model.Review: Which model fits your machine?Question 3 of 3
With llama3.2, what happens when you say “remember that my city is Porto Alegre”?
Correct
Since v2.0.1, the assistant directly saves requests like “remember that…”. llama3.2 still makes little use of the other tools.That’s not it
Since v2.0.1, the assistant directly saves requests like “remember that…”. llama3.2 still makes little use of the other tools.Review: Which model fits your machine?Test complete
You got {a} out of {n} right. Get them all right and the lesson will be marked complete.Choose an option. The explanation appears after your attempt.
07Takeaways from this lesson
A simple startPython 3.10+, setup, verification, and a conversation.
The model mattersllama3.2 remembers “remember that…” requests, but rarely calls the other tools. For files and commands, check what actually happened.