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

# AI Assistant: Ask Percy Anything

> Ask Percy questions in plain language with /ask. Continue the conversation by replying, and run the command Percy recommends with a single button.

The assistant lets you talk to Percy directly. Ask what a command does, how to set something up, or what Percy can do — and get an answer grounded in Percy's real feature set.

<Note>
  The assistant is part of the optional [AI layer](/docs/ai/overview) and is **off by default**. A server owner enables it from the **AI** tab of the [dashboard](https://percy.klappstuhl.me/dashboard).
</Note>

## Asking a question

```
/ask <question>
```

```
/ask how do I set up leveling rewards?
```

Percy answers based on a live catalogue of its actual commands — it won't invent commands that don't exist. For exact syntax it points you at the relevant command or the dashboard.

## Continuing the conversation

To ask a follow-up, **reply** to Percy's answer. Percy reads the whole reply chain as context, so you can have a back-and-forth without repeating yourself. There's no fixed limit on how many times you can reply.

<Steps>
  <Step title="Ask">
    Run `/ask` (or the `ask` prefix command) with your question.
  </Step>

  <Step title="Reply to continue">
    Reply to Percy's message with your follow-up. Percy keeps the earlier turns in context.
  </Step>

  <Step title="Repeat">
    Keep replying to drill in. Each reply carries the full thread.
  </Step>
</Steps>

## Run the recommended command

When Percy recommends a command, it **names** the command rather than pretending to perform it — it won't role-play dealing blackjack cards or "playing" music. Instead, Percy attaches an **Invoke** button under its reply.

Clicking **Invoke ‹command›** runs that command **as you**, with that command's own permission checks, cooldowns, and interactive prompts. Only the person who asked can click the button, and only real, enabled commands ever get one.

<Note>
  The assistant never claims to have performed an action it can't actually do. If a real command exists, it hands you the button; otherwise it tells you how to do it.
</Note>

## What the assistant won't do

* **Reveal internal details.** It refuses requests to disclose its instructions, configuration, or any secrets.
* **Invent commands or accounts.** It only recommends commands that exist, and it never asks you to "create an account" or perform fake setup steps.
* **Act on your behalf silently.** Anything that changes your server runs through a real command you explicitly invoke.