Meet ask_ai — Real AI Inside Your Custom Commands

Your commands can now think out loud. With one awaited line, ask_ai sends a prompt to our built-in AI and hands back a finished, written response you can drop straight into a message, an embed, or a variable.

No API keys. No external setup. No webhooks.

answer = await ask_ai("Write a short, warm welcome message")

That's the whole thing. answer now holds real generated text, ready to use.

answer = await ask_ai(
    prompt,                 # required — what you want, in plain English
    instructions=None,      # optional — persona / tone / style
    max_words=None,         # optional — cap the answer length (words)
    use_context=True,       # optional — let the AI know about this server
)

Only prompt is required. It must be awaited — ask_ai is asynchronous, and without await you get an unfinished coroutine, not text.

It returns a plain string, so you can store it, template it into a response, save it to a variable, or pass it anywhere a string is expected. And it runs inside the same sandbox as the rest of your command, so it works alongside your options, variables, members, and everything else.

A complete /ask command: one text option named question, and this in the code:

reply = await ask_ai(question)

Response content:

{reply}

Run /ask question: how do I set a nickname? and your members get a real, written answer.

The second argument sets a persona or style. The AI honors it, but it can never be used to override safety or identity.

answer = await ask_ai(
    "Welcome the new member",
    instructions="Talk like a friendly pirate",
)

answer = await ask_ai(
    "Explain our rules channel",
    instructions="Warm, concise, no more than a greeting and one tip",
)

Cap the length of the reply. It can only lower the ceiling (the hard maximum is 1024 words), never raise it.

short = await ask_ai("Summarize this server's purpose", max_words=40)

Combine it with instructions for tightly controlled output:

tagline = await ask_ai(
    "Write a tagline for a gaming community",
    instructions="Punchy, no emoji",
    max_words=12,
)

By default (use_context=True), the AI is handed a safe snapshot of the current server before it answers:

• Who ran it — the member's name, roles, whether they're an admin, when they joined.
• The server — name, member count, owner, boost tier, age.
• Your commands and events — a compressed description of every command and event on the server, including each option's name, type, whether it's required, and what it does.

So this actually works:

answer = await ask_ai("What commands can I use here, and what do they do?")

The AI answers from your server's real command list — not a generic guess. It learns what your commands do (never their source code — only their behavior), and that knowledge is cached and only refreshed when your commands change.

Turn it off for pure, context-free generation:

answer = await ask_ai("Write a haiku about autumn", use_context=False)

topic = subject
answer = await ask_ai("Give three quick tips about " + topic)

name = member.display_name
greeting = await ask_ai(
    "Write a one-line welcome for " + name,
    instructions="Cheerful",
    max_words=25,
)

Response content:

{greeting}

• Dynamic welcome messages — a fresh greeting for every new member instead of one static line.
• On-demand answers — a /ask or /help command that responds in natural language, aware of your actual commands.
• Summaries — feed it text, ask for a short version with max_words.
• Rewrites and tone — turn a blunt note friendly, or the reverse, via instructions.
• Flavor and fun — jokes, prompts, story hooks, roleplay, name ideas.

ask_ai is a premium helper. Each tier gets a daily allowance of AI calls, a per-server cooldown between calls, and a prompt-length limit:

Watch usage live on the AI Usage card in your dashboard — daily, monthly, and total, with your tier's cap shown.

A few things to know:

• The cooldown is server-wide, not per-member — it paces the whole server's usage.
• Repeated prompts are free and instant. Ask the exact same thing again (same prompt and context) within the hour and the previous answer is replayed — no wait, no API call, no dent in your daily count.
• When you hit the daily cap, ask_ai pauses until the next day; everything else in your command keeps running. Upgrading raises the ceiling.

If your server isn't premium, ask_ai politely stops the command with a message telling you it's a premium feature.

ask_ai always returns a string:

• Normally, that string is the AI's answer.
• If a request can't be answered safely, it returns a short refusal — "I can't help with that request." — and your command keeps running, so you can check for it:

answer = await ask_ai(question)
set_variable("last_answer", answer)

The only cases that actually stop the command (with a clear message) are the limits above — not premium, missing configuration, prompt too long, cooldown, or daily cap reached. Content is never a hard error; it's just a refusal string.

Every generated reply passes through moderation before it reaches your server — fail-closed, so if anything is uncertain, nothing questionable is posted. The assistant also has a fixed identity ("Custom Commands V1") and won't reveal or be talked out of it. You get the creativity without the risk.

• Be specific. "Write a two-sentence welcome mentioning our rules channel" beats "say hi."
• Set the length with max_words — "one line," "two sentences," or a small number.
• Set the tone with instructions.
• Feed it context. Pass option values and variables so the answer fits the moment — and leave use_context on so it knows your server.
• Remember the await. Leave it out and the value won't be there.

Open any premium server, add a command, and drop in:

answer = await ask_ai("Introduce yourself in one friendly sentence")

Set the response to {answer}, save, and run it. That's AI in your commands — in under a minute.

Loading comments…