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

# Crush (CLI/TUI Agent)

> Route your Crush terminal agent through LLM Router to automatically compress massive terminal logs, redact local secrets, and slash API costs.

***

> ## Documentation Index
>
> Fetch the complete documentation index at: [https://docs.llmrouter.app/llms.txt](https://docs.llmrouter.app/llms.txt)
> Use this file to discover all available pages before exploring further.

# Crush & LLM Router

[Crush](https://github.com/charmbracelet/crush) (or similar terminal-based AI agents) is a powerful CLI/TUI tool that handles code generation, debugging, and file operations directly inside your command line.

Because terminal agents frequently read massive error logs, command outputs, and local files, they consume huge amounts of input tokens. By routing Crush through **LLM Router**, you automatically apply **Middle-Out Context Compression** to bloated terminal outputs and ensure your local `.env` secrets are **Redacted** before they are sent to third-party AI providers.

<Warning>
  **Important:** When using LLM Router with custom CLI tools, you must configure
  the base URL to point to `https://api.llmrouter.app/v1` to ensure OpenAI SDK
  compatibility.
</Warning>

## Step 1: Installing Crush

If you haven't installed Crush yet, select the appropriate installation method based on your system:

<Tabs>
  <Tab title="Homebrew (macOS/Linux)">
    ```bash theme={null}
    brew install charmbracelet/tap/crush
    ```
  </Tab>

  <Tab title="NPM (Cross-Platform)">
    `bash npm install -g @charmland/crush `
  </Tab>

  <Tab title="Arch Linux">`bash yay -S crush-bin `</Tab>

  <Tab title="Nix">
    ```bash theme={null}
    nix run github:numtide/nix-ai-tools#crush
    ```
  </Tab>
</Tabs>

## Step 2: Modifying the Crush Configuration

To point Crush to LLM Router, you need to edit its local configuration file and add LLM Router as a custom provider.

### 1. Locate the Configuration File

Depending on your operating system, the `crush.json` configuration file can be found here:

<CodeGroup>
  ```bash macOS/Linux theme={null} theme={null}
  ~/.config/crush/crush.json
  ```

  ```powershell Windows theme={null} theme={null}
  %USERPROFILE%\.config\crush\crush.json
  ```
</CodeGroup>

### 2. Add LLM Router as a Provider

Open the `crush.json` file and add the LLM Router endpoint under your providers list. Make sure to replace `sk-router-your-api-key` with your actual LLM Router API key.

```json theme={null} theme={null}
{
  "providers": {
    "llmrouter": {
      "id": "llmrouter",
      "name": "LLM Router",
      "base_url": "https://api.llmrouter.app/v1",
      "api_key": "sk-router-your-api-key"
    }
  }
}
```

<Tip>
  **API Key Location:** You can generate your LLM Router API Key in the
  [Dashboard](https://llmrouter.app/dashboard). Your key is stored locally by
  Crush and is never exposed.
</Tip>

## Step 3: Launch Crush and Select Your Model

1. Run the `crush` command in your terminal to start the application:

```bash theme={null} theme={null}
crush
```

2. Open the command palette (usually `ctrl+p`) and choose **"Switch Model"**.

3. Because LLM Router acts as a universal gateway, you can type the name of *any* supported model, prefixed with the provider. For example:
   * `anthropic/claude-3-5-sonnet` (Best for heavy coding/refactoring)
   * `openai/gpt-4o`
   * `deepseek/deepseek-chat` (Highly cost-efficient)
   * `router-auto` (If you want LLM Router to dynamically pick the model based on your Tags)

## Step 4: The LLM Router Advantage in the Terminal

Once configured, you can use Crush as normal to generate code, execute terminal commands, and debug.

Because traffic is now flowing through LLM Router, you instantly gain:

* **Token Savings on Logs:** If Crush runs `npm run build` and ingests 3,000 lines of terminal output, LLM Router will dynamically compress the repetitive middle sections, saving you up to 80% on API costs.
* **Zero-Trust Security:** If Crush accidentally reads your `.npmrc` or AWS credentials from your local workspace, LLM Router's PII engine will redact them into `[TOKEN_REDACTED]` before sending the context to Claude or GPT-4.
* **Resilience:** If your primary AI provider experiences an outage, LLM Router will automatically fall back to your configured secondary provider without interrupting your CLI session.