> For the complete documentation index, see [llms.txt](https://ce-docs.keywordrush.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://ce-docs.keywordrush.com/ai-agents/safety-and-permissions.md).

# What your assistant can and can't do

Handing an AI assistant access to your website is a big deal, so Agent Access is built to keep you in control. Here's exactly how.

### It acts as you — and only as you

Your assistant can do only what the WordPress user you connected is allowed to do. It has no special powers of its own. If you connect a user who can't publish posts, the assistant can't publish posts. If you connect a user who can't change plugin settings, it can't change settings.

{% hint style="info" %}
**Want to limit what it can do?** Connect a more limited WordPress user. Create a dedicated user with just the role you're comfortable with (for example an Editor), generate the application password for that user, and the assistant inherits exactly those limits.
{% endhint %}

An **Editor** is the sweet spot for most people. It covers the entire content workflow — searching products, finding and editing posts, building pages from blocks, adding images and videos — while your plugin and module settings stay out of reach. Only connect an **administrator** if you actually want your assistant configuring modules, connecting shops, or changing Content Egg settings.

### Where your password ends up

An application password is a key to your site, so it's worth knowing where the one you hand over comes to rest:

* **Typed into a chat** — it stays in that conversation's history. Fine for a one-off session; change it afterwards.
* **Saved in a project's instructions or a custom GPT** — stored with your assistant and reused in every chat there. Convenient, but it sits there until you remove it.
* **Put in a config file on your own computer** (the MCP setup, or your own scripts) — it never enters a conversation at all. This is the most private option.

Whichever you use, the safeguards are the same: one named application password per assistant, so you can revoke just that one; a dedicated user, so the password is only worth what that user can do; and a fresh password whenever you think one has been seen by more people than you'd like. Generating a new one takes seconds.

### Your secrets stay hidden

Your affiliate keys, API keys, and other passwords are **never** shown to your assistant. When it reads your settings, secret values come back masked (shown as `••••` with only the last few characters). It can *update* a key if you ask it to, but it can never *read one back* — so a connected assistant can't leak your credentials.

### Drafts by default

When your assistant builds a page, it creates a **draft**. Publishing is a separate, deliberate step: the assistant only publishes when you ask it to, and only if the connected user has permission to publish. Nothing goes live behind your back.

### Destructive actions are labeled

Some actions change or remove existing content — removing products from a post, or replacing a page's content. These are marked as destructive, so a well-behaved assistant will confirm with you before doing them. If you're ever unsure, ask your assistant to explain what it's about to do before it does it.

### Everything is logged

Every action your assistant takes is recorded in **Recent activity** on the **Content Egg → AI Agents** page — which user, which action, when, and whether it succeeded. Nothing is hidden.

<figure><img src="/files/pWmPoSnHw3HAMrAuk9e5" alt=""><figcaption><p>The activity log on the AI Agents page</p></figcaption></figure>

### Turn it off anytime

You have two off switches:

* **Flip the master switch** — untick **Enable Agent Access** and Save. Every assistant loses access immediately.
* **Delete the application password** — go to **Users → Profile → Application Passwords** and revoke the one you created. That single assistant is cut off, while everything else keeps working.

{% hint style="warning" %}
Treat an application password like a key to your site. Use a **dedicated user** for your assistant, don't share the password, and revoke any application passwords you're no longer using.
{% endhint %}