Docs · Guide

miniQ on any model

miniQ is a desktop AI coworker with a miniq terminal client, both clients of one miniq-daemon. Its models come from one OpenAI- or Anthropic-compatible endpoint, the provider in settings.json. magpie makes that provider the local gateway, so miniQ runs on any model in magpie: DeepSeek, GLM, Kimi, Qwen, GPT or a ChatGPT subscription.

Set up miniQ

1. Add a provider in magpieAdd an API key on the Providers page, or sign in to a subscription.
2. Pick a model for miniQ on the Agents pageminiQ's row lists your current miniQ model and the models via magpie. Pick one of the latter.
3. Keep workingA running daemon is told the new provider, and open chats move onto the new model, with no restart.

magpie miniq deepseek/deepseek-v4-pro

On Windows, miniQ in a WSL distro is found too and set up the same way.

Where settings.json is

settings.json sits in the daemon's data folder, found as miniQ finds it. MINIQ_DATA_DIR comes first. Next is the folder named by miniq-data-dir.json beside the default, which exists when you moved the data folder in miniQ's Settings. Otherwise it is the default. macOS uses ~/.local/share/miniq too.

macOS / Linux: ~/.local/share/miniq/settings.json
Windows:       %LOCALAPPDATA%\miniq\settings.json
MINIQ_DATA_DIR set: $MINIQ_DATA_DIR/settings.json

What gets written

{
  "provider": {
    "baseUrl": "http://127.0.0.1:3425/v1",
    "apiKey": "",
    "model": "deepseek/deepseek-v4-pro",
    "apiProtocol": "chat_completions"
  }
}

baseUrl is the gateway's /v1. apiKey is magpie's own key for miniQ: miniQ sends no User-Agent of its own, so the key is how magpie knows the request is miniQ's. model is the model's catalog id. apiProtocol is chat_completions, because auto would guess Responses or Anthropic Messages from a model's name.

magpie changes only provider. It stashes your own provider whole. The rest of settings.json stays as it is: MCP servers, approvals, hooks and remote access. Where there is no file, magpie makes one with 0600 permissions, as miniQ does.

A running daemon

The daemon reads settings.json once, when it starts, and writes its own copy back on every settings change. So after writing the file, magpie also calls settings.update over the local WebSocket that daemon.json names, as miniQ's own Settings does. A daemon magpie can't reach keeps the old provider until it restarts. magpie then asks you to restart miniQ.

Switching back

Switching back puts your own provider back whole and tells the daemon. If you had no provider, magpie removes its own, and miniQ goes back to its default. If magpie made settings.json and nothing is left in it, the file goes too. You can also pick your own miniQ model in magpie; it goes into your restored provider.

magpie miniq default

FAQ

How do I use DeepSeek, GLM or a ChatGPT subscription in miniQ?

Pick a via-magpie model for miniQ on magpie's Agents page. magpie sets settings.json's provider to the local gateway's /v1 and tells the running miniq-daemon.

Do I need to restart miniQ?

Usually not. magpie tells a running daemon through settings.update, and open chats move onto the new model. Only a daemon magpie can't reach needs a restart, and magpie says so.

Why chat_completions and not auto?

auto guesses the protocol from a model's name and could pick Responses or Anthropic Messages. The gateway's /v1 speaks Chat Completions, so magpie names it.

Are my MCP servers and other settings changed?

No. magpie changes only settings.json's provider, and switching back restores the one you had.

More agents on the same models: Empryo on any model, Lody on any model and Cindy on any model. Something fails? See Fixing common errors, or ask on Discord.