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

# Safety and control

> What agents do on their own, what waits for you, and how to change or revoke access.

Giving an agent your email or your social accounts is a big step. This page explains what agents can do by themselves, what always comes back to you, and where to change your mind.

## What agents do on their own, and what waits for you

| Agents do this on their own | This comes back to you first |
| - | - |
| Read and search the apps you've given them access to | Sending, replying to, or forwarding an email |
| Research on the web | Creating, changing, or canceling a calendar event, and inviting people |
| Write drafts of emails, posts, and events | Publishing a social post |
| Organize your inbox: labels, archive, move to trash | |
| Build pages and dashboards in your project | |

What comes back to you arrives as an [approval card](/docs/features/approvals.md) in the conversation. You can edit it, approve it, or discard it.

* **Emails and calendar invites** always come back as a card in a conversation, even if you say "just send it".
* **Social posts** come back as a card by default. If you explicitly tell the agent to post right away, or you've told it before that it doesn't need to ask, it can publish directly.
* **Websites** go live as soon as you ask an agent to publish one. See [Pages](/docs/features/pages.md).

## Routines ask for permission up front

A [routine](/docs/features/routines.md) runs while you're away, so it can't stop to ask you. When an agent sets up a routine that will send email, permanently delete email, or change your calendar, it lists exactly what the routine needs and why, and waits for you to decide.

* You can approve everything, only some of it, or nothing. See [Permissions](/docs/features/approvals.md#permissions).
* Permission belongs to that one routine. It doesn't carry over to other routines.
* A routine waiting for your decision doesn't run. If it needs something you didn't approve, it can't do it.
* To stop a routine, pause or delete it from **Routines**.

## Choose which agents can use which apps

Connecting an app doesn't give every agent access to it. You decide per agent:

* **From the app**: open **Integrations**, select the app, and under **Agent access** select **Give access** or **Remove access** for each agent.
* **From the agent**: select **Connectors** on the agent's card to see and change what it can use.
* **Disconnect an app** from **Integrations** or **Settings → Connections**. Every agent loses access to it right away. See [Connect your apps](/docs/integrations/connect-apps.md).

## Memory you can see and delete

Agents remember facts and preferences you share so you don't have to repeat yourself. Open **Settings → Memory** to search, edit, or delete any memory. Memories can apply everywhere or to a single project. See [Memory](/docs/features/memory.md).

## In Slack

When an agent answers in a Slack channel, anyone in the channel can talk to it, so it doesn't use your connected accounts, such as email and calendar, or your private memories there. See [What stays in SureThing](/docs/integrations/slack.md#what-stays-in-surething).

## Spending stays in check

* A single run never spends more than the credits you had when it started, and runaway runs are stopped automatically.
* When your balance runs out, scheduled routines are skipped until credits are available again. Nothing runs up a bill.
* **Spending review** in the **Routines** view shows which routines cost the most over the last 30 days and suggests ways to save, such as pausing a routine or running it less often.

See [Plans and credits](/docs/plans-and-credits.md) for what typical work costs.

## Your data

* SureThing is SOC 2 Type 1 compliant. See the [Trust Center](https://trust.surething.io) for reports and security details.
* SureThing doesn't sell your personal information.
* Data from Google accounts is handled under Google's API Services User Data Policy, including its Limited Use requirements.
* To delete your account and everything in it, open **Settings**, go to **Danger Zone**, and select **Delete account**. This can't be undone.

For details on what's collected, which AI providers process it, and how long it's kept, read the [Privacy Policy](https://surething.io/privacy-policy).


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.