Build Discord commands by describing them.
Type what you want — "a /warn command that bans after 3 warnings" — and Nexus generates a production-ready slash command in seconds. No JSON to write, no code to deploy, no bot restart. This guide walks through how it works, with real prompt-to-command examples you can copy.
The bridge between "what you want" and "working code".
Most Discord bots give you two paths: rigid templates with toggles (MEE6, Dyno) or full custom code in Discord.js. Templates are fast but limited; code is unlimited but requires a developer. Nexus collapses both into a third path: you describe the behavior, the AI writes the JSON.
Behind the scenes, every Nexus command is a structured JSON IR (Intermediate Representation) that defines a trigger (slash command, message event, button click), optional preconditions (permissions, role checks, time gates), and a list of actions (send message, assign role, modify variable, call HTTP webhook). The bot runtime executes this IR deterministically at every event.
The AI Command Builder takes your prompt, picks the right IR shape from the DSL grammar, fills in the slots (channel IDs, role names, conditions), and ships a valid IR ready to deploy. No free-form code generation — the model outputs a constrained schema that the bot already knows how to run, so generation failures land as "couldn't understand your prompt" instead of broken commands.
From prompt to deployed in 5 steps.
The whole loop — describe, generate, review, deploy, iterate — runs in under a minute for most commands.
Describe the command in plain English
Open the AI Command Builder in your dashboard and type what you want as if you were briefing a junior moderator. The clearer your prompt, the better the result — but you don't need to know any technical terms.
Nexus generates the command IR
Within 3-5 seconds, the AI returns a structured JSON IR (Intermediate Representation) — a declarative blueprint of triggers, conditions, and actions. You see a summary card with the command name, what it does, and the permissions it requires.
Review and tweak (optional)
If something looks off — a missing condition, the wrong channel, an over-broad permission — open the JSON editor and fix the specific field. Or just send a follow-up prompt: "actually make it ephemeral" or "add a 30-second cooldown".
Deploy to your server
Click Deploy. The command goes live across all selected servers in under 5 seconds — no bot restart, no manual config push. Slash command appears in Discord immediately for everyone.
Test in Discord and iterate
Run the command in your test channel. If something's not right, go back to the chat and refine: "make the warning embed red instead of orange" or "let admins use it on themselves too". The next deploy is instant.
Six prompts that work today.
Each prompt below is a real command you can paste into the builder. Output is a working slash command, ready to deploy.
- Example 01 · Progressive discipline / moderation
/warnYour prompt"Create a /warn command that bans the user after 3 warnings, with the reason logged to #mod-log"
What Nexus generatesGenerates a slash command with mod-only permission, a required `user` and `reason` argument, a counter that persists across messages, and a conditional ban action when the count reaches 3 — plus an embed log to the channel you specified.
- Example 02 · Onboarding / verification
(automatic)Your prompt"Build a /welcome flow that DMs new members a custom message with their name and assigns them a Verified role after they react"
What Nexus generatesSets up a `guildMemberAdd` event listener that DMs the new member, posts a public welcome embed, then waits for a reaction emoji to grant the role. All wired in one IR — no separate setup steps.
- Example 03 · Self-serve roles with anti-raid gating
/roleYour prompt"Make a /role command for self-assigning the Gamer role, but only to members who joined more than 24 hours ago"
What Nexus generatesSlash command with a role select option, a precondition checking `member.joinedAt < now - 24h`, and an ephemeral confirmation reply. Includes the rejection message for users who joined too recently.
- Example 04 · Support ticket system
/ticketYour prompt"Build a /ticket command that creates a private channel for the user and pings the @Support role"
What Nexus generatesGenerates a slash command that creates a channel under a specified category, sets permission overwrites for the requesting user + the Support role, posts an opening message, and adds a close button via component interaction.
- Example 05 · Community engagement / contests
/giveawayYour prompt"Create a /giveaway command with a prize argument, duration in minutes, and only members with 100+ messages can enter"
What Nexus generatesSlash command with arguments for prize and duration, a reaction-based entry mechanism, a precondition that checks the user's message count from the leveling DB, and an automatic winner-pick at expiry.
- Example 06 · Reputation / kudos system
(automatic)Your prompt"When someone says \"thanks\" and mentions a user, add a +1 to that user's reputation"
What Nexus generatesGenerates a `messageCreate` event listener with a regex pattern for thanks-style phrases, mention extraction, and an increment to a per-user reputation counter stored in the bot's variable store. No /command needed — purely passive.
Want to see the actual generated JSON for these? Open the builder and paste any of the prompts above.
Where the AI shines — and where you'll want to tweak.
- Common patterns: warns, mutes, kicks, bans, welcome flows, ticket systems, role assignment, leveling commands
- Conditions you can phrase clearly: "only admins", "members joined > 24h ago", "in this channel only"
- Iterations: refining a command after first deploy ("make it ephemeral", "add a cooldown")
- Webhooks and HTTP calls when you specify the URL and the payload shape in the prompt
- Highly specific business logic the AI can't infer ("ban if score is above 80% but only on Tuesdays") — write those manually or describe them in 2-3 short prompts instead of one big one
- Complex multi-step flows with branches and loops — the AI handles 2-3 branches well, drops accuracy past that. Build incrementally.
- Channel and role names: prefer pasting IDs in the prompt when you have many channels with similar names — the AI might pick the wrong one
- Embed copy: the AI generates serviceable text, but if your server has a strong tone of voice, edit the embed body after generation
About the AI Builder.
The technical questions about how the AI Command Builder works. Need something specific? Ask in our Discord.
Nexus uses Anthropic's Claude as the language model behind the AI Command Builder. The model is given the full Nexus DSL grammar (action types, condition operators, valid argument schemas) so it generates IR that's syntactically valid by construction — not free-form text it then has to parse.
Generated commands pass through a Zod schema validation step before deployment, so syntactically broken commands never reach Discord. Semantic accuracy (does it do what you described?) is high for common patterns — moderation, welcome flows, role assignment, tickets — and lower for very specific or unusual logic. When in doubt, the AI usually asks a clarifying question rather than guessing.
No. The whole point is you don't. You describe what you want in any language Claude understands (English, French, Spanish, etc.), the AI generates the JSON IR, and the visual builder lets you tweak fields with dropdowns and inputs. JSON is exposed for power users who want it, but never required.
Free plan includes a daily quota of AI generations (resets every 24h UTC). Pro plan ships 50 AI credits per day, Agency 200. Each prompt-to-deploy cycle uses one credit. Editing an existing command via natural language ('make this ephemeral') also uses one credit.
Yes. Every command exposes a JSON view in the editor, and you can copy the full IR. The IR is portable across Nexus instances — useful if you want to migrate commands between servers, share a setup with other server owners, or version-control your command library in a Git repo.
No. The model only sees the prompt you type and the Nexus DSL grammar — never your server's messages, member data, or moderation history. We don't fine-tune on customer prompts and we don't share data between servers.
Your first command, in five minutes.
No credit card. Free tier covers 3 servers and a daily AI quota generous enough to bootstrap any community. Upgrade when you scale.