> ## Documentation Index
> Fetch the complete documentation index at: https://docs.voyager.so/llms.txt
> Use this file to discover all available pages before exploring further.
# Control approvals and spending
> Choose how Voyager answers questions, set spending preferences, and review requests for permission.
Choose how much Voyager should decide on its own, and when it should wait for you.
Creative questions, spending preferences and permission to run actions have separate controls. Choosing **Always ask** for creative questions does not change the agent's permission mode.
**Autopilot** takes the agent's recommended answer immediately, including approval for paid generation above your spend limit. Choose **Balanced** or **Always ask** when you want Voyager to wait before exceeding that limit.
## Choose how questions are answered
In the chat's message box, open **Question answers** and choose a mode.
| Mode | What happens |
| - | - |
| **Autopilot** | Takes the recommended answer immediately, including answers that approve paid generation above the spend limit. |
| **Balanced** | Counts down to the recommended answer, with a default of 60 seconds. Spend above your generation limit waits for you. This is the default mode. |
| **Always ask** | Waits for your answer every time. |
A question without a usable recommended answer still waits for you, including when the recommended answer needs you to type something.
Open **Settings → Creative preferences → Ask before exceeding**.
* **Generation spend limit** tells the agent when to ask about a paid generation batch. It starts at $5. Set it to $0 to ask before any paid generation in **Balanced** or **Always ask**.
* **Render time limit** tells the agent when to ask about a local render expected to take longer. It starts at 10 minutes.
Voyager asks before spending more than your limit (in **Balanced** and **Always ask**). The spend limit is a preference the agent follows, not a hard spending cap or an account balance. An estimate is not a guaranteed final charge.
Check what the recommendation will approve. For generated video, this can include the stills, the shots to animate, and an estimated cost and time.
In **Balanced**, click **Wait** to stop the countdown or **Choose now** to accept the recommendation immediately. You can choose another option instead. A question asking to approve paid work without an estimate also waits for you.
## Know which account pays
Your selected chat account pays for the agent's chat usage. Generated images, video, music and voice use a generation account separately. Your Claude or ChatGPT does not cover those media charges.
You can use your own Claude or ChatGPT and your own media keys after signing in and finishing setup. A paid Voyager plan is optional. For Voyager credits and plan options, see [plans and credits](/account/plans-and-credits) and [current pricing](https://voyager.so/pricing).
**Automatic fallback** is on by default. If a generation provider is out of credits, rejects its sign-in or is down, Voyager tries the next one in your list. That may bill a different account. It may also adjust settings slightly to match what the next provider supports.
Open **Settings → Accounts → Generation providers** to review the order or turn **Automatic fallback** off. You can also ask the agent to pin a provider for a particular job, for example: “Use only my FAL account for this generation; stop if it cannot run.” See [generation providers](/account/generation-providers).
## Choose permissions for commands and file changes
Open **Settings → General → Permissions**. Changes apply from the next message; projects can override the default.
| Mode | What to expect |
| - | - |
| **Supervised** | Asks before commands and file changes. |
| **Auto-accept edits** | Approves edits automatically and asks before other actions. |
| **Auto** | Supported agents approve routine actions; others still ask. |
| **Full access** | Routine commands and edits run without prompts. This is the default. |
These modes concern actions the agent takes. They do not replace **Question answers**, your spending preferences, or your computer's own permissions. On macOS, **Settings → Permissions** is a different page for access such as screen recording; see [Mac permissions](/setup/mac-permissions).
When an approval card appears, read the action. **Allow** permits it; **Deny** refuses it. **More approval options** can include **Allow for this session** and **Deny and stop**. The available choices can vary by agent.
## Check your settings
Return to the chat and confirm the **Question answers** mode. Check your saved limits before starting a paid batch. If you want to review every creative choice and paid-work question yourself, use **Always ask**.
## Common problems
Check for **Autopilot**, or a completed **Balanced** countdown. Switch to **Always ask** for future questions. Changing the mode does not undo work already approved or submitted.
Autopilot immediately accepts the recommended answer, including above-limit spending. Switch to **Balanced** or **Always ask**. The spend limit guides when the agent asks; it is not a hard account cap.
**Always ask** controls question cards. To request approval for commands and file changes, choose **Supervised** under **Settings → General → Permissions**.
Review **Generation providers** and **Automatic fallback** in **Settings → Accounts**. Disable fallback or pin the intended provider for future work. Ask the agent to identify which account handled the existing job before trying it again.
Review the look and cost before animating shots.
Handle failed, queued or uncertain work without paying twice.
This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.