Models

mini-AlbertCode works with the model you connect: a hosted provider, any OpenAI-compatible service, or a model server on your own machine.

Connecting a provider

The list is grouped:

One OpenAI-compatible service is connected at a time: connecting DeepSeek replaces Groq, and the other way round.

Each provider is tagged as the service last found it: connected, key not accepted or not answering. A provider is called connected only once it has accepted your key, or once its server has answered with a model; a saved key alone is not enough.

Then, as the provider needs:

• Its address, for a local server or another service. The scheme is there already, so you type only the rest: https:// for a service on the internet, http:// for one on your machine or your network (in the browser it changes by itself as you type such an address, until you choose one). Ollama is 127.0.0.1:11434/v1, LM Studio 127.0.0.1:1234/v1. A whole address pasted, scheme and all, keeps its own scheme.
• Its API key, typed into a field that doesn't show it. An empty key changes nothing.

The key is checked with the provider as soon as it is typed, before anything is saved: a wrong key is said at once and never stored. Then the models it can use are listed, to choose one; it is used at once and becomes the provider's model. In the browser, Advanced: a different model for each tier (folded) gives each of the provider's three tiers (Advanced, Balanced, Economy) a model of its own, each one you can switch to from the model list; unless set, all three use the model chosen.

Connect DeepSeek
  Address   [https://] api.deepseek.com
  API key   ••••••••••••••••          ✓ Accepted
  Model · 2 models this key can use
            (•) deepseek-flash   ( ) deepseek-v4-pro
  > Advanced: a different model for each tier
                                     [Test] [Save and use]

Saved keys are kept by your system's own protection (the macOS Keychain, Windows data protection, or your Linux desktop keyring) and never shown again. See Security.

Choosing a model

Change the model at any time; the next message uses it.

The model list has the models you used recently first, then each connected provider's models as the provider lists them, and Connect a provider… at its foot; in the browser the providers not connected yet are folded under Not connected. The one in use is ticked. A model's ID that is not listed can be typed: the list offers Use it with each connected provider (compatible:my-model). Choosing one sets it for this session; ctrl+s in the terminal's list, or Also the default for new sessions in the browser's, also makes it the default for new sessions. The choice is shared: a model chosen in the browser is the terminal's, and the other way round.

Under the browser's box, the provider menu lists the connected providers (choosing one moves to the model you last used with it, or its first) and the model menu lists that provider's models.

When no model is ready, both interfaces say so and open the list instead of sending your message; what you typed waits in the box.

Checking

albert-mini models              # the providers, each connected or what it needs, and the models ready
albert-mini models test         # one tool-calling request to the chosen model: can it do the work?
albert-mini models test openrouter:some/model

Local models

Start your model server (Ollama, LM Studio, llama.cpp's server), then connect Local server and give its address. Nothing leaves your machine. A local model must support tool calling to make changes; the models test command shows whether it does.