Lesson 6 · 15 min
The AI model is the easiest part to swap. In the cloud, OpenRouter gives you access to more than a hundred models with one key. On your computer, Ollama runs at no cost per message and sends nothing outside your machine. The choice isn't a matter of taste: it's a table with speed, quality, and cost.
Start lesson
Goal: compare one cloud model and one local model using the same question, then decide what to use each for based on your numbers.
Click each block.
Cloud for power, local for privacy. You can use both.
Click each step.
Choose a question you would actually ask the assistant, preferably one related to a task in your meu-jarvis.md.
Use exactly the same text with both models.
Organize these into three priorities: finish the sales report, reply to two clients, and prepare for Thursday’s meeting.
Time it from sending until the full response arrives. Run it twice: the first call to the local model is usually slower because it needs to load.
Write down the second time.
Time in seconds, cloud and local.
Read both responses without knowing which is which, if you can ask someone to switch their order. Rate each from 1 to 5: Was it accurate? Did it follow your style? Was it useful?
A short note for each response is enough.
Would I use this response without rewriting it? Yes = 4 or 5.
For the cloud model, check the OpenRouter dashboard to see how much the question cost. Locally, each message costs nothing; the cost is keeping the computer on.
Write your decision in one line: which model for what.
Daily routine locally; client messages in the cloud.
The cloud key follows the same rule as the token: only in .env or the vault.
OPENROUTER_API_KEY=<paste the key you created at openrouter.ai/keys>
"providers": {
"default": "openrouter",
"openrouter": { "model": "anthropic/claude-sonnet-4-20250514" },
"ollama": { "model": "llama3.2", "base_url": "http://localhost:11434" }
}Switch the default provider to ollama (model llama3.2), keep openrouter as an alternative, and run the tests again. Don’t change anything outside the configuration.
Question: | Model | Time (s) | Rating 1-5 | Cost | |--------|----------|------------|------| | Cloud: | | | | | Local: | | | 0 | Decision: I’ll use ____ for ____ and ____ for ____.
Exercise E5 in sequence 2.
Run the same question through a cloud model (OpenRouter) and a local one (Ollama), if your computer can handle it. Fill in the minha-comparacao.md table.
Done when: you can say which model to use for what, using your own numbers.
Check that the OpenRouter key is only in .env (or in the vault) and that .env does not appear in git status.
Done when: A search for the key outside .env finds nothing.
The explanation appears after your answer.
Question 1 of 3
When does a local model (Ollama) make sense?competency
Question 2 of 3
You switched from OpenRouter to Ollama. What needs to change in the channel and memory?
Question 3 of 3
Why run the question twice on the local model before recording the time?
Choose an option. The explanation appears after your attempt.