Help CenterGetting startedInstall & sign in to agents
Install & sign in to agents
Getting started

Install & sign in to agents

How to install and sign in to the agents on your computer, like Claude Code or Codex, and how QU Studio uses your agents.

Q is the default name; you can rename it on its page, and the app uses the new name everywhere.

What to install

Q and the working sessions run on the agent programs on your computer. Two are mainly supported; any one is enough:

AgentOfficial install guideSign-in command
Claude CodeClaude Code setupclaude auth login
Codex CLICodex CLIcodex login

Q can only run on Claude Code or Codex. Other ACP agents (such as Gemini CLI or OpenCode) are listed under Settings › Agent › More agents; they can run sessions, but can't take tasks from Q or call in helpers.

Install and sign in

  1. Install it following the official guide. QU Studio doesn't install it for you or ship an installer.
  2. Sign in from a terminal: run the sign-in command above and sign in with your own account. A subscription (Claude Pro / Max, a ChatGPT plan) works, and so does your own API key; follow each provider's docs.
  3. Check in QU Studio: in Settings › Agent, click "Check again" and look for "Signed in". A row that isn't installed has "How to install"; one that isn't signed in has "Copy sign-in command". Paste the copied command into a terminal and run it.

If you use several computers (see Multiple computers), each one needs its own install and sign-in. The Computers page shows which agents each one has and whether they're signed in.

Which model new sessions use

  • Settings › Agent › Model presets: each preset sets the provider, model, effort, fast and permissions. Sessions you open and sessions Q opens start with the one marked "Default", and you can still change it when you open one.
  • Each preset can have a "When to use", like "small edits and adding tests" or "when Claude's quota is running low". Q picks a preset by it when handing out work, and so do sessions calling in helpers; a preset without one is never picked on its own.
  • If a preset names a model this computer doesn't have, the default model is used instead, without an error.
  • Presets override the default model you set in Claude Code itself.
  • Where the other models are set: see Usage & limits · Where to change models.

Use other providers' models (DeepSeek, Kimi, GLM…)

DeepSeek, Kimi, GLM, Qwen and MiniMax all offer Anthropic-compatible endpoints, so Claude Code can call them. The engine is still Claude Code, so Q's assignments, steering and taking over in a terminal all work as usual. Give a model preset an "engine profile" and only sessions opened with that preset switch over; other sessions are untouched.

  1. Write your own Claude Code settings file, for example ~/.claude/deepseek.json, with the endpoint, key and model names in env (follow the provider's docs):

``json { "env": { "ANTHROPIC_BASE_URL": "https://api.deepseek.com/anthropic", "ANTHROPIC_AUTH_TOKEN": "your DeepSeek API key", "ANTHROPIC_DEFAULT_OPUS_MODEL": "deepseek-chat", "ANTHROPIC_DEFAULT_SONNET_MODEL": "deepseek-chat", "ANTHROPIC_DEFAULT_HAIKU_MODEL": "deepseek-chat" } } ``

  1. In Settings › Agent › Model presets, add a preset: Claude Code, model Opus (mapped to the provider's model by the lines above), and put the file's full path in "Engine profile".
  2. Pick this preset when you open a session; to let Q use it for assignments, say when in "When to use".
  • QU Studio only hands the path to Claude Code. It never reads the file or touches your key, and the key can live in your own environment variables instead.
  • Codex works too: define [model_providers.<name>] in ~/.codex/config.toml and put that name in the preset's "Engine profile".
  • Costs for these sessions are computed at Anthropic / OpenAI prices and would be wrong, so only tokens are recorded, with no dollar amount. The provider's quota can't be read either.
  • The preset only applies on the computer where you set it; sessions opened on another computer don't carry the profile.
  • Sessions with an engine profile don't use fast mode (other providers' models don't have it). If the preset doesn't pick a permission mode, they start in "ask every time" rather than following permissions.defaultMode in the file.
  • Don't put disableAllHooks in that file: it would also turn off QU's safety hooks.

How we use your agent

QU Studio only runs the agents you installed and signed in to yourself on your computer, like Claude Code or Codex, with your own account. Specifically:

  • Detect and run, nothing more: detection only checks that the program is there (--version) and whether it's signed in (through the agent's own official command or interface). Then it runs the agent on your computer when you ask.
  • No installing, bundling or modifying: it uses the copy you installed, as published. It never removes or restricts any sign-in method the agent offers.
  • Never reads your account files: QU Studio doesn't read, copy or store credential files such as ~/.claude or ~/.codex/auth.json, and never handles your passwords or tokens.
  • No sign-in inside the app: QU Studio has no Claude or ChatGPT sign-in. Signing in only happens in the agent's own flow; if an agent isn't signed in, QU Studio just tells you to sign in from a terminal.
  • Usage is yours: usage is billed directly to your own Claude / ChatGPT account or API key. QU Studio charges nothing, and doesn't pay for, resell or route your usage.
  • Your environment stays yours: QU Studio doesn't change your network, proxy or account settings. When something can't connect, it tells you why and leaves the fix to you.

Claude Code is a product of Anthropic, and Codex is a product of OpenAI. QU Studio is an independent product and isn't affiliated with or endorsed by either.

Keep an eye on usage

  • Subscription limits are designed for ordinary individual use. Q can open several sessions at once, and sessions can call in helpers, so over long stretches your quota goes faster than when you work alone. You can set "Max tasks at once" per computer on the Computers page.
  • A few features make extra model calls on your usage; each can be turned off or pointed at a cheaper model: - Guess my next message, Name tasks automatically: turn them off in Settings › Agent, and change the model under "Model for next-line guesses and task names"; a fast, cheap model is the default. - Tidy memory when idle: turn it off in Settings › Memory & privacy, and change the model under "Model for tidying memory"; at most once a day. With pay-as-you-go sign-ins such as an API key, remaining quota can't be read, so it won't tidy automatically.
  • The bottom left shows how much of each account's limit is left on this computer; the Usage page breaks spend down by day, project and model.
  • When one provider's quota is running low, the session offers to continue on the other one. You can also pick the other provider's model in the model picker at any time; it switches from your next message and brings the conversation so far. See Sessions.

Common questions

It's installed, but QU Studio says "Not found" QU Studio reads the PATH from your login shell and also checks a few common install locations. Open a new terminal window and run claude --version or codex --version to make sure it works. If it does and QU Studio still says not found, click "Check again" in Settings › Agent.

Several copies of Codex are installed QU Studio uses the newest one (including the copy that comes with the ChatGPT app). If the one in your terminal is older, Settings › Agent tells you.

It says "Sign-in status unknown" When the status can't be read, try starting a session. If it really isn't signed in, the agent reports an error in the session; sign in from a terminal and try again.

Do I sign in to ChatGPT inside the app to use Codex? No, and there's no place to do so. Run codex login in a terminal.

Do I need a subscription? Not necessarily. Claude Code and Codex both support signing in with your own API key; set it up following each provider's docs.