Models and providers

The "Models" group in the settings, section "Models and providers", decides what the agent runs on: a Claude subscription, a cloud with an API key (DeepSeek, OpenRouter, Kimi, Z.ai, MiniMax) or your own server — llama-server, llama-swap, LM Studio, anything that serves /v1/messages over the Anthropic protocol.

The provider list lives on the server: added from the laptop, the phone sees it right away.

Without a Claude subscription: a cloud with an API key

Clauder doesn't need a Claude subscription. Any cloud that speaks the Anthropic protocol will do — you pay it for what you use, and Claude Code on the machine still runs the session, it just talks to someone other than Anthropic.

1. "Settings → Models" → "Connect models", pick a cloud: DeepSeek, OpenRouter, Kimi, Z.ai (GLM) or MiniMax.
2. Paste the API key. Where to create one is linked right in the form; the address and the model list are already filled in, there is nothing to type by hand.
3. "Check and connect".

Models are taken from the provider itself, and if it doesn't serve a list — from Clauder's preset. The key is stored only on the server machine, in Clauder's database, and goes only to the cloud itself.

The check tells you what to do instead of "error 401": the key didn't fit — where to get a new one; the account ran out of money — where to top it up; too many requests — wait a minute.

If there is no Claude login at all, Claude models in the list are labelled "needs login" — not hidden, you can log in any time — and new chats go to the connected cloud on their own. The second setup step, a banner on the home screen and the "Not logged in" error in the chat all lead to connecting with a key.

Android has a short dialog for this: the cloud, the key, one button — checking, connecting and switching new chats to that cloud in one go. It works with the server on the phone itself too: the Clauder Runtime companion has a "No Claude subscription — a cloud with an API key" card, and the cloud is connected to the server living on the phone — DeepSeek right from the phone, no home machine needed.

What new chats run on

A switch at the top of the section: Claude subscription or any connected cloud with a model to pick. It is the same default model as in the general settings for new chats — only the model changes, effort and the rest stay. Projects with their own settings and chats already open are left alone; any chat's model can be changed in the chat itself.

When a cloud is the default, its models come first in the chat's model picker, and Claude models go as a group below.

Marking models

OpenRouter has hundreds of models, and a drop-down of all of them would be a scroll. So in the provider card models are found by search and marked with a tick, and only marked ones make it into the chat's model picker. On connecting, the model the check passed on is marked straight away.

For the other clouds and your own servers it works the other way round: while nothing is marked, the chat shows every model; mark a few — only they remain.

Your own server

"Own server" in the same form — the address and the key (if the server runs with one), on/off, refreshing the model list and a connection check.

A provider's model appears in the common picker as a separate group under the server's name, as a link like llamacpp/gpt-oss-20b. An unloaded one is labelled: the first reply on it will come after the weights are fetched.

There is no separate provider switch — choosing a model is choosing a server.

Two quirks that can't be worked around

Switching to a provider's model restarts the session. Claude Code has no API of its own for "change provider": the connection is the environment of the CLI's child process, and that is set at start.

Such models have no effort levels. effort is a property of Anthropic's models; someone else's server simply doesn't have it.

The connection check

It goes step by step, and every failed step gives a hint:

1. the address answers;
2. the model list;
3. the Anthropic protocol — a 404 means an OpenAI-compatible server;
4. tool schemas — llama.cpp trips over heavy ones, and such a tool is simply switched off for the session;
5. a streamed reply.

On an unloaded model the check takes up to a minute, and that is normal. A cloud answers within a few seconds.