Tutorial 23 · Providers in depth: setting up each model

What the video says

1. Providers is where you choose the models JMAX thinks with. It lives in the menu, under Management and administration. Normally the administrator sets it up once, on the day you go live.
2. The screen has two parts. At the top, Roles: which model does what, with five jobs. Below, the list of providers and their models.
3. Each role is a row with three things: the provider, the model and the Save button. The change applies instantly, without restarting JMAX, and the top of the window shows which model ended up active.
4. The big brain reasons, summarizes and plans. It handles the hard requests, so this is where your highest quality model goes, even if it takes a little longer to answer.
5. The fast brain carries the conversation and the voice. Here speed matters, not size: pick a small, nimble model so answers come back without a wait.
6. The fallback steps in only when the main one doesn't answer. Always point it at a different provider from the big brain: if your own server goes down, the cloud fallback saves the day.
7. Vision is the model that looks: screenshots, photos and scanned documents. It has to be a model that understands images, like Gemma three.
8. Embeddings is the search across your documents: it turns each text into numbers so it can be found by meaning. The usual choice is JMAX built-in, which runs inside the application, with no server and no key. Kimi, Anthropic and DeepSeek don't offer this function, which is why they don't appear here.
9. Below are the twenty-six providers. The ones tagged local run on your computer or on your network and ask for no key. The rest are internet APIs and need one.
10. Click the name to open the card. On a local one you only need to check the endpoint, which is the address of the machine, and tick the active box. The models you see are the ones that machine has right now.
11. On an internet provider, the get one link opens the page where the key is created. You paste it, press Save key, and JMAX encrypts it on your computer: it is never shown again and never leaves, except towards that provider.
12. Discover models asks the provider for its catalogue and stores it. Test connection sends one short question and tells you whether it answers. You can test with the key you just typed, without saving it first.
13. Test every model asks each model a question and builds a table: how long it takes, whether it is fast or slow, whether it accepts tools and what it is good for. That is how you find out which model on your server is worth using.
14. On every card you decide what happens to personal data going to that provider: never send it, replace it with placeholders, or send it as is. Health, passwords and cards never leave, whatever you choose.
15. Some providers have a free tier: Groq, Cerebras, Mistral, OpenRouter or GitHub Models. The key is obtained with an account and no card, and they work very well as a fallback.
16. Watch out for gateways like FreeTokens: they listen on this computer but forward to the internet. JMAX treats them as an external provider, so the personal data filter applies to them too.