> ## Documentation Index
> Fetch the complete documentation index at: https://docs.mossx.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Detect and install

> Detect local CLIs, confirm the install plan, and fix PATH and custom-path issues.

Open **Settings → CLI configuration** and pick an engine on the left. The app detects the local binary and offers **Install** / **Update**. Clicking those shows a plan first, then runs after you confirm. **It will not silently run an install script for you.**

Credentials, providers, and capability differences live on each engine page. This page covers detect, install entries, and troubleshooting.

<Columns cols={2}>
  <Card title="Claude Code" href="/en/desktop/providers/claude">
    Hosted providers, local `settings.json`, CLI login.
  </Card>

  <Card title="Codex" href="/en/desktop/providers/codex">
    `~/.codex/` and sandbox.
  </Card>

  <Card title="Kimi" href="/en/desktop/providers/kimi">
    `~/.kimi-code/config.toml`.
  </Card>

  <Card title="Grok" href="/en/desktop/providers/grok">
    `~/.grok/config.toml`; use WSL on Windows.
  </Card>

  <Card title="OpenCode" href="/en/desktop/providers/opencode">
    `opencode.json`; env injection does not change the local file.
  </Card>

  <Card title="Pi" href="/en/desktop/providers/pi">
    `auth.json` and `/login`.
  </Card>

  <Card title="DeepSeek Harness" href="/en/desktop/providers/dsh">
    host / port and Web UI.
  </Card>

  <Card title="Provider overview" href="/en/desktop/providers/overview">
    How to choose, capability differences, where config is written.
  </Card>
</Columns>

Gemini, Cursor, Copilot, Qwen, and similar items appear under **Coming soon**. This page does not document install steps for them.

## Workflow

<Steps>
  <Step title="Pick an engine">
    Open **Settings → CLI configuration** and open the engine you want to install.
  </Step>

  <Step title="Install">
    Click **Install** and confirm the plan, or install yourself with the terminal commands on that engine page.
  </Step>

  <Step title="Confirm the command works">
    Open a new terminal and run the matching `--version`, such as `claude --version`, `codex --version`, or `kimi --version`.
  </Step>

  <Step title="Add credentials">
    Go back to that engine page: local config, a hosted provider, Pi `auth.json`, or the DSH Web UI.
  </Step>

  <Step title="Switch in the input">
    When status is ready, you can ask a question.
  </Step>
</Steps>

## What the in-app install preview shows

When you click **Install** in the app, the preview command depends on the engine:

| Engine                             | Preview command type      |
| ---------------------------------- | ------------------------- |
| Claude / Grok                      | Official install script   |
| Codex / Kimi / OpenCode / Pi / DSH | `npm install -g …@latest` |

Grok has no npm release. In-app **Install** for Grok is not available on Windows; run the official script in WSL. OpenCode and DSH have no in-app uninstall, so you do not delete login data or `$DSH_HOME` by accident.

Exact commands are in the **Install** section on each engine page.

## When detect fails

1. Open a new terminal and confirm the binary is on `PATH`
2. A GUI opened from a launcher sometimes cannot see your shell `PATH`. Write it into system environment variables and restart the app
3. Override a custom CLI path on that engine page
4. Come back to Settings and detect again
5. Confirm the engine is not in the **Disabled** group
6. For DSH, also confirm the host is up: open `http://127.0.0.1:3080` in a browser