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

# Provider Setup

> Connect your AI providers, configure defaults, and tune speed vs quality.

## Authentication

Each provider has a different auth method. During alpha, Claude and Codex are the supported providers. CCR + OpenRouter is managed by Whim and will not require setup once it launches.

### CCR + OpenRouter

<Note>Coming soon. CCR + OpenRouter is currently disabled during alpha.</Note>

No setup required. CCR is managed entirely by Whim. When it launches, API usage is billed as part of your CU allocation.

### Claude Subscription

<Steps>
  <Step title="Get your OAuth token">
    Run this in your local terminal (requires [Claude
    Code](https://docs.anthropic.com/en/docs/claude-code/overview) installed):

    ```bash theme={null}
    claude setup-token
    ```

    Copy the output token.
  </Step>

  <Step title="Paste in Whim">
    Go to **Settings > My Defaults > Claude** and paste the token. Click
    **Save**.
  </Step>

  <Step title="Verify connection">
    The status indicator shows a green dot with **Connected** when ready.
  </Step>
</Steps>

<Warning>
  Your OAuth token is encrypted before storage and never exposed in logs or to
  other workspace members.
</Warning>

**Token management:**

* The settings panel shows when your token was created and last used
* Paste a new token to replace the existing one
* Clear the field and save to disconnect

### Codex Subscription

<Steps>
  <Step title="Start the login flow">
    Go to **Settings > My Defaults > Codex** and click **Sign in to ChatGPT**.
  </Step>

  <Step title="Enter the device code">
    Whim displays a device code and a link to the OpenAI verification page. Open
    the link in your browser.
  </Step>

  <Step title="Authorize">
    Enter the code on the OpenAI page and approve. The page confirms when
    complete.
  </Step>

  <Step title="Verify connection">
    The connection status updates to **Connected** once verified.
  </Step>
</Steps>

<Note>
  Device codes expire quickly. If it expires before you finish, click **Sign in
  to ChatGPT** again.
</Note>

## Connection status

| Indicator                    | Meaning                 |
| ---------------------------- | ----------------------- |
| Green dot - **Connected**    | Authenticated and ready |
| Grey dot - **Not connected** | Needs setup             |
| Spinner - **Checking**       | Verifying connection    |

You can only set a provider as your default if it shows **Connected** and is available to your workspace.

<Note>
  CCR + OpenRouter is platform-managed, so it currently appears connected.
  During alpha it is still unavailable and cannot be selected.
</Note>

## Workspace and user defaults

### Workspace defaults (admin)

Go to **Settings > Workspace > Defaults** to configure for all members:

* Default provider and model
* Permission mode
* Fast mode
* Reasoning effort

### User defaults

Go to **Settings > My Defaults** to override workspace defaults for your tasks only. User defaults take precedence over workspace defaults. Per-task settings override both.

### Provider-specific settings

| Setting          | CCR + OpenRouter | Claude                    | Codex                       |
| ---------------- | ---------------- | ------------------------- | --------------------------- |
| Model            | Coming soon      | Yes                       | Yes                         |
| Permission mode  | Coming soon      | Yes                       | Yes                         |
| Fast mode        | No               | Yes                       | Yes (native GPT models)     |
| Reasoning effort | No               | Yes (Low/Medium/High/Max) | Yes (Low/Medium/High/xhigh) |

***

## Fast mode

Fast mode tells the provider to prioritize response speed. Useful for straightforward tasks - simple edits, boilerplate, quick fixes.

| Provider   | Effect                                      | Supported models  |
| ---------- | ------------------------------------------- | ----------------- |
| **Claude** | Enables fast output mode in Claude settings | All Claude models |
| **Codex**  | Sets `service_tier=fast`                    | Native GPT models |
| **CCR**    | Not available                               | -                 |

<Note>
  Fast mode for Claude uses the same model and does not change reasoning effort.
  Use **Reasoning effort** below to control Claude's `--effort` level.
</Note>

Enable fast mode at three levels:

1. **Per task** - toggle when creating a task
2. **User default** - **Settings > My Defaults**
3. **Workspace default** - **Settings > Workspace > Defaults**

<Info>
  Codex has fast mode **enabled by default** for GPT 5.6 Sol. Claude has fast mode
  **disabled by default**.
</Info>

## Reasoning effort

<Info>
  Reasoning effort is available for both Claude and Codex. Fast mode is a
  separate control.
</Info>

### Claude

| Level      | Best for                                                                   | Supported models                              |
| ---------- | -------------------------------------------------------------------------- | --------------------------------------------- |
| **Low**    | Simple, well-defined tasks                                                 | All Claude models                             |
| **Medium** | General coding work                                                        | All Claude models                             |
| **High**   | Complex tasks needing deeper analysis                                      | All Claude models                             |
| **xhigh**  | Difficult problems - complex debugging, architecture, multi-file refactors | Claude Opus 4.8 and Claude Opus 4.8 (1M) only |
| **Max**    | Extended thinking on the hardest problems                                  | Claude Opus 4.8 and Claude Opus 4.8 (1M) only |

<Note>
  Claude reasoning effort controls the `--effort` flag passed to the Claude CLI.
  When unset, Claude uses its own default behavior.
</Note>

### Codex

| Level      | Best for                                                                   | Supported models |
| ---------- | -------------------------------------------------------------------------- | ---------------- |
| **Low**    | Simple, well-defined tasks                                                 | All Codex models |
| **Medium** | Moderate tasks, some reasoning helpful                                     | All Codex models |
| **High**   | Complex tasks benefiting from deeper analysis                              | All Codex models |
| **xhigh**  | Difficult problems - complex debugging, architecture, multi-file refactors | All Codex models |

Set reasoning effort at the same three levels as fast mode: per task, user default, or workspace default.

## Speed vs quality reference

<Info>
  For Claude, fast mode and reasoning effort are separate controls. For Codex,
  fast mode is separate and applies only to native GPT models.
</Info>

| Configuration                                 | Speed            | Quality       | CU impact            |
| --------------------------------------------- | ---------------- | ------------- | -------------------- |
| Claude Haiku 4.5 + fast mode                  | Fastest (Claude) | Good          | Least container time |
| Claude Sonnet 4.6 + high effort               | Moderate         | High          | Moderate             |
| Claude Opus 4.8 + max effort                  | Slowest          | Highest       | Most container time  |
| Codex GPT 5.6 Sol + fast mode + low reasoning | Fastest (Codex)  | Lower         | Least container time |
| Codex GPT 5.6 Sol + xhigh reasoning           | Slower           | Highest (GPT) | More container time  |

Fast mode and lower reasoning effort reduce agent time per task, directly lowering CU costs. During alpha, supported providers only consume CU through container time.

## Troubleshooting

<AccordionGroup>
  <Accordion title="Claude token shows 'Not connected' after saving">
    Your OAuth token may have expired. Run `claude setup-token` locally to
    generate a fresh token and paste it in Whim.
  </Accordion>

  <Accordion title="Codex device code expired">
    Device codes are time-limited. Click **Sign in to ChatGPT** again to
    generate a new one.
  </Accordion>

  <Accordion title="Tasks fail immediately after switching providers">
    Verify the provider is connected (green dot) before setting it as your
    default. Existing tasks keep the provider they were started with.
  </Accordion>

  <Accordion title="My user defaults aren't taking effect">
    User defaults only apply to new tasks. Check **Settings > My Defaults** (not
    workspace defaults) and verify the provider is connected.
  </Accordion>
</AccordionGroup>