How to Set Up a Discord AI Bot? Gemini Chat Step by Step

The assistant starts off and depends on your plan

• The assistant starts disabled. AI settings are turned off in a new server. Any member using /ai before toggling the switch will see a "disabled" warning.
• Access depends on your plan. If the tab appears locked or you get an access error when toggling the switch, your server's plan may not include this feature. Check the pricing page for details.
• The bot must have permission to write in the target channel. If the response cannot be delivered, no error message appears on the screen; the bot will simply appear to stay silent. If you are unsure about channel permissions, check out the role and permission system guide.

This topic in VeyroBot:Discord AI Bot →

Step 1: Turn on the assistant

1. Open the Dashboard → your server → AI Studio tab.
2. Turn on the AI Assistant switch at the top of the tab. This switch is linked to the Google Gemini AI module switch on the Overview page; changing one will also update the other.
3. After configuring the settings below, click Save AI Settings.

Step 2: Choose a personality

The personality determines the bot's conversational style. There are four options:

Step 3: Extra rule and response length

• Custom System Prompt: A server rule added on top of your chosen personality; it does not replace the personality. It can be up to 1,000 characters long. Example: "Address members as 'Captain' and keep your answers short."
• Response Length Limit: Can be set between 500 and 8,000 in increments of 250; the default is 4,000. A lower value means shorter and faster responses.

Answers exceeding Discord's single message limit are not cut off; they are split and sent across up to five messages. A note indicating that it has been shortened is added to the end of any response longer than that.

Step 4: Restrict chat to a single channel

You can select a single text or announcement channel from the Allowed AI Chat Channel field. If left blank, the assistant will work in all channels. When a request comes from somewhere else after a channel is selected, it is handled as follows:

How do members talk to the bot?

• /ai question:<your message> — works only in servers; the response is visible to everyone in the channel.
• @VeyroBot <your message> — the mention must be at the very beginning of the message. If you just type and send a mention, the bot will remind you how to talk to it.
• Reply — if you use Discord's Reply feature on any message sent by the bot, your text goes straight to the AI. This is the most natural way to keep a conversation going.

The response language comes from the Bot Language setting under General Server Settings. If the setting is set to "Automatic", members using /ai receive responses in their own Discord language. Responses do not mention anyone: even if the model's text includes @everyone, a role, or a user mention, no one receives a notification.

How does memory work?

With every response, the bot sends the last 12 memory logs of its conversation with you on that server to the model. Since questions and answers are stored as separate entries, this roughly translates to the last 6 Q&As; each entry is stored with a maximum of 1,000 characters. All three methods use the exact same memory.

• Memory is server-wide: it is shared across channels within the same server and does not carry over to another server.
• AI conversations of other members are not appended to your response; a chat in a staff channel never leaks into an answer in a general channel.
• While responding, the bot knows the server name, the name of the channel you are in, and your display name.

If the bot does not reply or replies unexpectedly

• Placing the mention in the middle of a sentence. "Hey @VeyroBot how are you" will not trigger it. The mention must be at the very beginning.
• Mentioning the bot's role. The bot's role often shares the same name as the bot, and both might appear side by side in the mention list. Mentioning the role does not start a conversation; select the bot itself from the list.
• Clashing with a custom command. If the text following the mention matches a custom command trigger, the custom command will respond instead, and the AI will not kick in. A short trigger set up with Contains in Sentence matching can intercept many questions.
• Replying to the bot outside the restricted channel. No warning appears in this scenario; the bot might seem broken.
• Enabling the profane personality in a public server. The extra rule does not replace the personality; it builds on top of it. To change the tone, change the personality itself.

Is AI right when answers must be accurate?

Suitable for: communities where members ask questions, seek coding or translation help, or enjoy chatting with a character; those who want to set up an entertainment channel with a gamer personality in gaming servers.

Not suitable for: support desks that require strictly accurate factual information. The AI can generate incorrect answers and, unless specified in your custom instructions, does not know your server rules or announcements. Custom commands are much safer for questions that require fixed answers.

Frequently asked questions

Does the bot respond to every message in the channel?

No. It only replies to the /ai command, messages starting with a mention of the bot, and replies given to one of the bot's messages.

Does the AI remember my conversations in other servers?

No. Memory is kept on a per-server and per-user basis; it is shared across different channels of the same server, but never shared across servers.

What happens if many members ask questions at the same time?

Requests are queued and processed one by one. The queue rotates fairly among servers; even if one server sends too many requests back-to-back, requests from other servers will be interleaved. During peak times, responses may be delayed by a few seconds. If a Gemini model fails to respond, the bot tries the next model in line.

Can it generate inappropriate content?

With every request, the model is instructed not to generate sexual or NSFW content, and to politely refuse such requests—this rule takes precedence over the personality and the extra server rule. This is an instruction and not an absolute guarantee for the model output. Out of the four personalities, only the blunt one contains profanity in its instructions.

For a general introduction to the feature, check the Discord AI bot page; for all settings, refer to the Gemini AI and custom commands documentation; and for command details, check out the ai card.

Last updated: September 26, 2026. The trigger methods, personalities, limits, and memory behaviors in this article have been verified against VeyroBot's current version of the AI command, message handler, and dashboard settings validation.