How to build a Discord bot with AI
People mean two different things by a Discord bot with AI. Some want a bot that talks like ChatGPT in a channel. Others want an AI to build the bot for them, so that describing a giveaway is enough to get one. Both work on SCNX. They are set up in different places and they cost different things.
Updated 18. november 202611 min read
On this page
The short version
- A bot that chats is a module: you switch it on, point it at a channel and pay one AI Coin per answer.
- To have features built, you connect your AI assistant to your server once and describe what you want.
- What it builds starts switched off. You test it and switch it on yourself.
What you need
- Your own bot on SCNX. Creating one takes about two minutes
- For building: Custom Commands V3 switched on. It is in beta and open to every server
- For building: an AI app that supports custom connectors, such as ChatGPT, Claude or Gemini
Two meanings of "AI bot"
In the first, AI is a feature of the bot. Members write in a channel and the bot answers in sentences. Setting that up is a matter of minutes, and every answer costs a little.
In the second, AI is the builder. You tell an assistant what your server needs, and it puts together commands, buttons and scheduled posts that then run without any AI involved. A giveaway built this way needs no AI Coins.
A bot that chats in a channel
Switch on the module
Open Discord-Bot, then Modules, switch on AI-Chat-Channel and save.
Pick the channel
Open the module's configuration and add an AI-Channel. In the field AI-Chat-Channel you select the channel the bot should answer in. Create a new channel for this, so the bot doesn't reply to every line in #general.
Set a slowmode in Discord
Edit the channel in Discord and set a slowmode of about 30 seconds. Every answer costs AI Coins, and without a slowmode one enthusiastic member can use up the month's allowance in an evening.
Save and reload
Save the configuration and use Reload configuration on the bot's status card. Then write something in the channel.
- Personality
- How the bot talks. "Helpful Chat-Bot (default)" suits most servers. "Funny friend" and "Rude coffee shop employee" are role-play, and their answers need a closer eye from your moderators.
- Response Length
- Short (1-2 sentences), Medium or Long. Short answers keep a busy channel readable.
- Trigger Mode
- "All messages" answers everything written in the channel. "Mentions & replies to bot only" lets people talk to each other and call the bot when they want it, which also saves AI Coins. The keyword modes answer only when a message starts with your Trigger Keyword, for example "Hey Bot".
- Custom System Prompt
- Your own instructions, up to 1000 characters. They replace the Personality setting. The prompt is checked automatically when your bot starts, and a prompt that breaks our terms is not used. After changing it, type
=== RESET ===in the channel so the bot forgets the old conversation. - Answer as a custom persona?
- The channel answers under its own Persona name and Persona avatar instead of your bot's. For this the bot needs the Manage Webhooks permission in that channel. If the persona can't be used, the bot answers under its own name.
Illustration. The real interface may look different.
An answered message costs one AI Coin, and you are not charged when generating an answer fails. Every server receives AI Coins each month, our paid plans include more, and you can top up on your server's pricing page in the dashboard.
Two things to know before you switch it on. The messages in that channel are sent to OpenAI to generate the answers, so tell your members. And AI models get things wrong with great confidence, so someone on your team should keep an eye on the channel.
Images with /imagine
The second AI module is called /imagine. Once it is switched on, members type /imagine with a description and the bot posts the generated image. An optional size picks square, landscape or portrait.
An image costs between 10 and 25 AI Coins, depending on its size and on the quality you set in the module's configuration. Turn on Enable Rate-Limit? there as well. It limits how many images one member can request in a time frame you choose.
Before you build: switch on Custom Commands V3
What your AI assistant builds are flows of Custom Commands V3. V3 is in beta, and the server owner or a co-owner switches it on once.
Open the page
Go to scnx.app/ccv3-beta and pick your server, or open Custom Commands under Discord-Bot in the dashboard.
Read what you are taking on
Flows run with your bot's full Discord permissions. Anyone who may edit custom commands in your dashboard can build flows that use them, so check your trusted admins first.
Confirm and enable
Tick the two boxes and click Enable Custom Commands V3.
Switch the bot to the beta version if asked
V3 runs on the beta version of the bot. If your bot isn't on it, the page offers Switch to beta and enable V3. Nothing you have set up is lost, but you can't switch back to the stable version afterwards.
Restart the bot
Click Restart bot now. V3 is active once the bot is back.

Screenshot from our test server. Your dashboard may look different.
Custom commands you built before keep running and stay editable. If V3 isn't for you, Switch back to V2 on the V3 overview stops your flows again. The move to the beta version of the bot stays.
Connect your AI assistant
You connect the AI app you already use to your server. The connection is set up once and takes about a minute.
Copy the connector URL
In your server's dashboard, open AI assistant in the sidebar. Your connector URL is at the top, with a Copy button.
Add it to your AI app
In Claude, open Settings, then Connectors, then "Add custom connector". In ChatGPT, open Settings, then Apps and Connectors, turn on developer mode under the advanced settings and create a connector. In Gemini, open Settings, then Connected apps, and connect a custom MCP server. The dashboard page has a tab with the steps for each app.
Sign in and choose what it may do
Your AI app sends you to SCNX. Pick the servers. For each area you choose Read only, Read and make changes or no access. Under Access expires you set 30 days, 90 days, one year or never. Then click Approve access.
Start a chat
Open a new chat in your AI app and describe what you want.
- Who can connect
- The server owner, or a trusted admin who is allowed to change the bot's configuration in the dashboard.
- The bot has to be online
- Your assistant reaches your server through your bot. While the bot is stopped, its requests fail.
- Not sure yet
- Start with Read only. Your assistant can look at your flows and test them, but it can't change anything. You can reconnect with more access later.

Screenshot from our test server. Your dashboard may look different.
Some AI apps only allow custom connectors on a paid subscription. That is between you and the app's maker. SCNX does not charge for the connection.
Prompts that work
Write the way you would explain it to a moderator. Your assistant can see the names of your channels and roles, so call them by name. It can't see your members' messages.
| You ask | You get |
|---|---|
| Build me a level system with role rewards every 10 levels. | A rank command, levels saved per member and a flow that hands out the roles. |
| Add a ticket panel with a button that opens a private thread. | A panel message with its button and the flow that opens the thread, as a draft for you to look over. |
| Run a giveaway in #events every Friday at six. Members join with a button, one winner. | A scheduled post with a Join button, the saved entries and the draw. |
| Why did my welcome flow fail yesterday? | It reads the run history and names the step that failed and what that step was trying to do. |
| Starboard should need 5 stars instead of 3, and show me what you changed. | The one setting it changed, plus a saved copy from before so you can undo it in one step. |
- Say who, where and when
- "Every Friday at six in #events" builds faster than "a regular giveaway". The same goes for roles: name the role that may use a command.
- Ask for a plan first
- For anything bigger, have it describe what members will see and which channels and roles it will use before it builds. Correcting a plan is quicker than correcting a flow.
- Ask for a test run
- A test run goes through the flow step by step and sends nothing. Read it before you say yes.
- Change one thing at a time
- Ask what it changed after each request. Every change can be undone, and small changes are easier to undo than big ones.
A first request to copy
Look at my Discord server and suggest three features it is missing. Then build the one I pick as a draft, and show me a test run before anything goes live.
You decide what goes live
Everything your assistant builds starts as a draft that is switched off. A test run shows what each step would do without sending a message or changing anything on your server. You switch it on when you are happy with it. Flows that ban members, delete channels or do something similarly risky always arrive switched off, whatever you told the assistant.
Before your assistant changes something that exists, the previous version is saved, and you can go back in one step. The AI assistant page in the dashboard lists every change under Recent changes and shows how much of this hour's allowance is used.
The connection only reaches the servers you picked and ends on the date you set. You can revoke it earlier on the same page. Your assistant can't start or stop your bot, change its token, or touch backups, analytics and billing. It only changes the settings of built-in modules such as welcome messages or levels if you allowed the Bot modules area, and it asks you before those changes go live.
Illustration. The real interface may look different.
Beta, plans and what it costs
Building this way is part of Custom Commands V3, which is in beta. Servers use it every day, but it still changes and you may run into bugs. During the beta our support covers bugs you can reproduce. It can't help with building. The beta page lists all of this before you confirm.
SCNX doesn't charge for the connection. Your server's plan sets how many changes your assistant can make per hour, and higher plans allow more. Custom Commands V3 itself runs on every plan, including our free Starter plan. The subscription for your AI app, if it needs one, is separate.
When something doesn't work
- Your AI app has no place to add a connector
- Custom connectors are a paid feature in some apps, and ChatGPT hides them until developer mode is on. Check the tab for your app on the AI assistant page.
- "You don't have any eligible servers"
- You are logged in with an account that is neither the owner nor a trusted admin with the right permission. Ask the owner to connect, or to give you the permission to change the bot's configuration.
- The assistant says it can't change anything
- The connection is Read only. Connect again from your AI app and choose Read and make changes for the area you need.
- Requests to your server fail
- Your bot is probably stopped. Start it on its status card and try again.
- It built the flow, but nothing happens in Discord
- The flow is still a draft. Tell your assistant to switch it on, or switch it on yourself in the V3 overview.
- It stops in the middle of building
- This hour's allowance is used up. The AI assistant page shows the usage and when it resets.
- The AI chat channel stays silent
- If the bot says the server is out of AI Coins, top up or wait for the next month. Otherwise check the Trigger Mode, and that the bot may see the channel and send messages in it.
Having AI write bot code instead
An AI assistant can also write you a bot in discord.js or discord.py. You get a file of code. Running it is your job: a machine that stays on, a token kept out of the code, and a fix every time a permission is missing or the library changes. Our guide on hosting a bot 24/7 shows what that takes.
That route is a good one if you want to learn how bots work, or if you need something no platform offers. If you are already on it and tired of restarting the bot, we wrote down how that setup compares with building on SCNX.
Common questions
- Which AI apps work?
- ChatGPT, Claude, Gemini and other apps that support custom connectors. The dashboard has the steps for each.
- Does building with AI cost extra?
- SCNX doesn't charge for the connection. Your plan sets how much your assistant can change per hour, and higher plans allow more. Some AI apps want a paid subscription for custom connectors.
- What does the AI chat channel cost?
- One AI Coin per answered message. Every server gets AI Coins each month, our paid plans include more, and you can top up in the dashboard. A feature your assistant built runs without AI, so a giveaway or a level system costs no AI Coins.
- Can my AI assistant read my members' messages?
- No. It sees the structure of your server: channel names, role names and your bot's permissions. Message content and member data are not part of the connection.
- Can it break my server?
- It builds drafts that stay off until you switch them on, risky flows always arrive switched off, and each change it makes can be undone. Look over what it built before you switch it on.
- Do I have to learn the editor?
- No. You can build most things in the chat and open the visual editor when you want a closer look.
- Who can connect an assistant?
- The server owner, or an admin who is allowed to change your bot's settings in the dashboard. Switching on Custom Commands V3 itself is up to the owner and co-owners.
- Which AI answers in the chat channel?
- The answers are generated by OpenAI's models. The messages of that channel are sent there, together with your server's ID and name. Messages from your other channels are not.
Start with a free bot
Log in with Discord, add your server and create your bot. The dashboard walks you through each step.
Get started for freeFree on our Starter plan. A short ad video every 14 days keeps the bot online.