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

# Prowl AI Features: Chat, Image Generation & Summaries

> Prowl's AI feature brings GPT-powered chat, image generation, and thread summarization to your Discord server — configurable per channel.

Prowl's AI feature brings a capable, context-aware assistant directly into your Discord server. Powered by configurable GPT models, it can hold conversations and generate images from text prompts — all without leaving Discord.

<Info>
  AI features are enabled by default. Server admins can configure the model, system prompt, and other settings from the dashboard.
</Info>

## AI Chat

You can have a full conversation with Prowl's AI in any channel where AI is enabled.

```text theme={null}
/ai chat <prompt>
```

Prowl maintains short-term context within a conversation, storing the last 20 messages per user per server. You can ask follow-up questions without repeating yourself.

```text theme={null}
/ai chat Explain the difference between roles and permissions in Discord.
> [Prowl responds]
/ai chat Can you give me an example using a moderator role?
> [Prowl responds with context from the previous answer]
```

To clear your conversation history and start fresh:

```text theme={null}
/ai clear
```

<Note>
  AI chat context resets when the bot restarts or when significant time passes between messages.
</Note>

## Image Generation

Prowl can generate original images from a plain-English description using the `/ai imagine` command.

```text theme={null}
/ai imagine <prompt>
```

**Example prompts:**

```text theme={null}
/ai imagine a neon-lit cyberpunk city at midnight, rain-slicked streets, ultra-detailed
/ai imagine a cozy medieval library with floating candles and stacked bookshelves, warm tones
/ai imagine the Prowl logo as a stained glass window
```

Prowl will process your prompt and post the generated image directly in the channel. Generation uses DALL-E and typically takes a few seconds.

<Tip>
  More descriptive prompts produce better results. Include details like style ("oil painting", "pixel art", "photorealistic"), lighting, mood, and subject matter.
</Tip>

## AI Configuration

Server admins can configure AI settings from the dashboard or via slash commands.

| Command               | Description                                   | Permission    |
| --------------------- | --------------------------------------------- | ------------- |
| `/ai config`          | View the current AI configuration             | Manage Server |
| `/ai model <model>`   | Set the AI model                              | Manage Server |
| `/ai prompt <prompt>` | Set the system prompt that shapes AI behavior | Manage Server |

### Available Models

| Model           | Description                        |
| --------------- | ---------------------------------- |
| `gpt-3.5-turbo` | Fast and cost-effective (default)  |
| `gpt-4`         | More capable, slower               |
| `gpt-4-turbo`   | Improved GPT-4 with larger context |
| `gpt-4o`        | Latest multimodal model            |
| `gpt-4o-mini`   | Lightweight GPT-4o variant         |

### System Prompt

You can set a custom system prompt to shape how the AI behaves in your server. The default is:

```text theme={null}
You are a helpful Discord bot named Prowl. Be concise and friendly.
```

Change it to fit your community — for example: `"You are a friendly assistant for a gaming community. Keep responses appropriate for all ages."`

### API Key

Prowl resolves API keys in this order:

1. Per-guild key set in the dashboard
2. Global key stored in the bot's database
3. Environment variable (`OPENAI_API_KEY` or equivalent)

<Note>
  When using the bot owner's API key, rate limits apply: a 5-second global cooldown and a 60-second per-user cooldown. Guilds with their own API key are not rate-limited.
</Note>

<Warning>
  AI-generated content is not infallible. Always review the AI's outputs in sensitive contexts and encourage your members to treat AI responses as helpful suggestions, not authoritative facts.
</Warning>