# How the agent replies

# How the agent replies

The agent replies with one image when a whitelisted account posts the agent's @handle and a character trigger word, each as a separate word.

## What triggers a reply

Two things must be true. The author is on the whitelist. The post contains the agent's @handle and a character trigger word, each as a separate word.

Matching ignores case. The words can appear anywhere in the post. Type the trigger word without a slash. The rest of the post becomes the image prompt.

## Examples

*The agent @pepebot with the trigger word `pepe`*

| Input | Output |
| --- | --- |
| `@pepebot pepe on a skateboard` from a whitelisted account | One image reply |
| `pepe on a skateboard @pepebot` from a whitelisted account | One image reply |
| `@pepebot, pepe on a skateboard` | No reply. The comma is attached to the handle. |
| `@pepebot on a skateboard` | No reply. No trigger word. |
| `@pepebot /pepe on a skateboard` | No reply. The slash is part of the word. |
| `@pepebot pepe on a skateboard` from an account not on the whitelist | No reply |

## What the agent posts

The agent replies to the triggering post with one image. The reply has no text.

- Replies and quote posts that mention the handle count.
- Direct messages are not read.
- The agent's own posts are ignored.
- With several characters, the first matching character in the agent's list wins, not the first word in the post.
- If generation fails, nothing is posted and the sender gets no message.
- Before the character finishes training, no reply is posted.

## Character changes

The agent uses the characters of its matching Telegram bot, up to 4. Edit them in the **Meme Bot** tab with the switch set to **Twitter**.

A change reaches the running agent the next time you save the whitelist.

- [The whitelist](/x-agents/xagent-whitelist)
- [The trigger word](/generation/generation-how#the-trigger-word)
- [What an X agent is](/x-agents/xagent-what-it-is)