Build with Q

The Quick Setup Composer (internally known as Projects) allows you to build and configure your server using natural language. Instead of clicking through dozens of individual settings pages, you tell Q what kind of community you want, and Q generates a complete implementation plan.

You can find this builder by navigating to Projects in the main sidebar, clicking + Create Project, and selecting the Build with Q tab. Alternatively, you can always open the floating crew chat widget on any dashboard page.

Quick walkthrough

1. Navigate to Projects → + Create Project → Build with Q tab.
2. Type your prompt in the chat, for example: Build me a FIFA tournament community with a rules channel, a #matches channel, an announcements automation for match results, and a Verified role for new members.
3. Q will analyze your request and generate a structured blueprint outlining the exact roles, channels, module configurations, and automation rules it will create.
4. Review the proposed plan and click Apply. Every card Q puts in front of you says what state it is in above its title: Ready to apply with the number of changes while it waits for you, then Applying, then Applied. A card that says Replaced was superseded by a newer suggestion and can no longer be applied, and Not applied means an approved change did not go through.
5. Q will execute the changes in one shot. Once completed, you can run a Test Project simulation to verify that all automation rules and permissions work correctly.

Jobs Q runs end to end

Some requests are a whole job rather than a question: build me a post, set my server up, put a ticket system in, draw a matching set of pictures. Q follows a written procedure for those, and you see it as a checklist that fills in as each point is finished.

• A finished post. Q agrees the style and the format with you, looks up anything that changes (weather, news, prices) on the web instead of answering from memory, offers to draw the picture or take one from your library, and only then puts the card in front of you, already complete.
• A server setup. One card of questions, a look at what the server already has, then a single proposal for you to approve.
• A ticket system. Permissions first, then a draft of the first panel, then the proposal. Nothing reaches Discord until you approve it.
• An artwork pack. One style agreed once, then an icon, a banner and an emoji drawn to match it.

A point is ticked when the thing it names has actually happened, not when Q says it has. If a required point is still missing when the answer ends, Q is sent back to finish it before you see the result.

Light and dark

The chat has two full themes and its own control for them, in the chat header: tap it to move between System, Light and Dark. There is no theme setting on a settings page, because there does not need to be one.

The choice follows you. Pick light on your phone at /m and the desktop panel opens light too, on that browser. System means the operating system on the phone screen, and dark in the dashboard panel, which is what the panel is floating over.

Following what Q is doing

While Q is part-way through something, a strip sits above the box you type in. It carries the goal in the words you used, where the work stands, the last few things Q did, and what happens next. Open Show the record for the full list with times. It is a summary of real work, never a view of Q thinking out loud, so there are no tool names in it and no percentage: a number nothing measures would only look like progress.

The strip goes away when the work is finished. What is unfinished stays: close the tab, come back tomorrow, and the same state and the same question are waiting, because the record lives on the server and not in your browser.

Anything still waiting on you is also listed under Needs you on the Overview, with a link straight back into the conversation it belongs to. A decision you left open is findable from there without hunting through your chats.

If you decide you do not want a piece of work after all, Stop this at the foot of the strip ends it. Whatever was waiting to be applied stops being applicable, including on a card still open in another window, and it stops appearing under Needs you.

The language you write in

Write to Q in any of the nine languages Qreate speaks and the whole chat follows, including Simplified Chinese and Traditional Chinese for Taiwan. The header, buttons, card labels and activity strip switch with the conversation, whatever language your dashboard uses. A short Chinese reply made only of shared characters keeps the writing system already selected instead of switching scripts.

It ignores what it cannot read. A bare ok, a yes or an emoji leaves the chat in the language it was already in, so one short reply in the middle of a long conversation never flips everything back. Spoken messages are decided from the audio itself, which is steadier than reading a transcript that recognition may have mangled.

Seeing what Q did

While Q works, a line above its answer names the step it is on. Open it and you get the whole list in order: the settings it read and how many rows were there, any search it ran and how many sources came back, a picture it drew, and each change it applied. Every step carries how long it took.

A step that failed stays in that list, marked as a failure, with the reason the platform gave. That is the part worth reading when something did not turn out the way the answer describes it: the reason there is the real one. The list stays with the answer after the turn finishes, so you can go back to it later.

The chat header's Changes button lists what this conversation applied, with one Undo for each Apply. When Q reads analytics, its stats card shows the same numbers as Overview and lets you switch between 7, 30 and 90 days.

What people actually type

Q reads plain sentences, not command syntax. You don't need to know a module name or a setting key, just say what you want the way you'd say it to a person. Some real shapes of a request:

• Building something from scratch: "Set up a ticket system with a support category and a log channel only staff can see."
• Changing a setting that already exists: "Turn off auto-kick in verification, we keep losing legit members before they finish."
• Checking what's already there: "What reaction role panels do we have right now?"
• Drafting a post: "Write an announcement for #general, the giveaway ends tonight at 9pm."
• Artwork: "Draw a server banner, dark blue with a wolf howling at the moon."
• Artwork with your own mascot: "Put our mascot on a welcome banner." Q draws from the images in your QAssets whose name or tag says logo, mascot or brand, or from the file you name, and keeps them as they are.
• Checking the web: "Did Discord change anything about bot verification requirements recently?"
• Removing something: "Delete the #archive-2024 category, we're done with it."

Key capabilities

• Structure scan: Q automatically scans your server's real channel names, roles, and member count so you don't have to explain your layout. It targets existing channels directly and avoids duplicating them.
• Turnkey builds: A single sentence can scaffold roles, channels, and rules, resolving IDs automatically.
• Artwork on request: Q draws the pictures too. Ask for a server icon, a banner, an emoji, a role icon or the image on top of a post, pick from the options, and it lands in your QAssets library at the right size.
• Idea generation: Stuck on what to build? Click the Surprise me chip to receive 2-3 concrete, creative ideas suited for your server's size and profile.
• Publish-ready copy: Any text Q drafts (rules, welcome messages, announcement events) is output as fully-formatted, ready-to-post Discord markdown, never generic placeholders.
• Persistent memory: Q keeps the durable facts about your server across sessions so you never repeat yourself. The brain icon above the chat box opens everything it keeps: edit the wording, delete an entry, add something it never picked up, or switch memory off entirely.

What Q remembers, and how to change it

Memory is why Q stops asking the same questions. It keeps two separate things: a server profile every admin here benefits from (what the community is, your tone, what you dislike), and personal notes from your own conversations on this server.

Open the brain icon above the chat box to see all of it. Every entry is editable, because correcting a memory beats deleting it: delete a half-wrong note and Q learns the same wrong thing again from the same conversation. You can also write an entry yourself, without working it into a chat first.

• Remember this server: the shared profile. Switch it off and Q stores nothing new about the community, and reads none of what is already there.
• Remember me: your own notes on this server. Your call alone.
• Switching a scope off leaves existing entries in place, so turning it back on restores what Q knew instead of starting from zero.
• While a scope is off, Q says so instead of promising to remember.

"The same one as yesterday"

You can point back at earlier days without re-explaining. Q carries a short summary of your previous sessions on this server: when each one happened, what you opened it with, and which changes were really applied while it was open, with the panel's real name. So "reuse the reaction roles panel, same as yesterday" resolves to the actual panel instead of a question.

• What it has: the date, the title of the session, the last thing you asked, and what got applied. Enough to find the thing you mean.
• What it does not have: the messages themselves. Q cannot quote a past conversation and will tell you so rather than pretending.
• Only applied changes count as fact. A session where you discussed a panel but never pressed Apply is not a panel, and Q says that instead of naming one that does not exist.
• Recent days stay visible. A busy morning does not push yesterday out of reach: the summary spreads across days rather than keeping only the newest few chats.

Seeing what Q is doing while it works

When Q needs to look at what you already have, it reads it while it writes rather than afterwards, and the status line above the reply says what it read and what came back: "Checked Reaction Roles, 6 in total", "Checked Tickets, nothing there yet". Each line appears once the read is done, so it is a report and not a promise.

This is also why answers about your existing setup come back faster. If your message names something Q will need to look at, it is already being read before Q starts writing, and Q does not spend a second pass fetching what it was handed.

How much Q is allowed to do on its own

There is a switch for this, and it is yours. Open it from the shield button in the chat header, or from Q's authority in the dashboard. Four settings, from one where Q only explains things and builds nothing, through the default where it prepares everything and waits for you to press Apply, up to one where it gets on with routine work by itself. If a whole setting is too blunt, the fine-tuning underneath changes one capability without touching the rest.

• Only the server owner can raise it. Anyone with access can lower it, because pulling Q in should never wait for someone else to log in.
• Deleting is the exception. A channel or a role always waits for you to confirm, at every setting, including the most autonomous one.
• So is acting on a member. A ban, a kick or a timeout always waits for a person, at every setting. A rule that would do any of those automatically waits too: writing one is writing a standing instruction to act on members.
• Nothing is hidden from you. Whatever Q applies on its own shows up in the chat as done, with the record of what changed. If a module kept only part of what you asked for, Q says which part it did not keep instead of calling the whole thing done.

Ask Q to remove a channel or a role

"Delete the #old-announcements channel, we don't use it anymore" works the same as any other request, but this is the one thing Q never finishes for you, at any authority setting. By default it won't even offer to: deleting is switched off until an admin turns it on.

1. Turn deleting on from Q's authority, either by picking Trusted or Admin or by turning on just the destroy capability on its own, leaving everything else where it is.
2. Ask for the removal by name, the way you'd say it to a person: "remove the #announcements-old channel" or "take away the Muted role, nobody uses it".
3. Q shows you a card built from the real object on Discord right now, not from what it remembers: the actual name, how many channels sit inside it if it's a category, how many members hold it if it's a role.
4. Read what goes with it before you decide. A category takes every channel inside it along with it. A role is worn by real members, and they lose it the moment it's gone.
5. Confirm. A bigger deletion (a category with channels in it, a role a lot of people hold) asks you to type the exact name before the button turns on.

Ask Q to take down a message it posted

Panels, announcements and anything else Qreate put in a channel can be removed the same way: "take down that old reaction roles panel in #roles". Q shows you the real message it is about to remove, read from the channel right now, and waits for you to confirm. This runs on the revise capability in Q's authority, separately from deleting channels, so you can allow one without the other.

• Only Qreate's own messages. A message from a member, a moderator or another bot is not something Q can touch, whatever permissions the bot has. Those you remove yourself: hover the message in Discord, open the ⋯ menu, pick Delete Message.
• Deleting a panel already does this. Removing a reaction roles, rules or ticket panel takes its message down with it, so you rarely need to ask for the message separately. It is there for a leftover message with no panel behind it.
• It does not come back. Discord keeps no copy, and anything attached to the message, including buttons people were using, goes with it.

Ask Q to deal with one member

"Time out that guy for ten minutes", "ban the account spamming invite links in #general", "lift the timeout on Sam". Q can ban, kick, time out, warn, lift a timeout and lift a ban. It never does any of it on its own, at any authority level: you get a card with the actual account on it, read from Discord right then, and you press the button. Your name is what goes in the moderation log.

1. Say what should happen and to whom. If Q does not already know which account you mean, it shows a member search and you tap the person, so you never have to go and find an ID.
2. Q asks how long a timeout should last if you did not say. It does not pick a length for you.
3. Read the card. It shows the avatar, the account name, when they joined, their top role, and what the action actually does to them.
4. Press it. The action runs under your name, with the reason you gave, and it appears in Moderation with the rest of your cases.

• One person at a time. Q will not bundle a ban in with a settings change, so you can never approve one by accident while accepting the other.
• It tells you when Discord will refuse. The server owner, Qreate itself, and anyone whose highest role sits at or above Qreate's are all off limits, and the card says so instead of letting you press a button that fails.
• A pattern is not a person. "Stop the raid" or "handle the link spam" is automod, raid guard and the escalation ladder, and Q will set those up instead. Ask it.

Ask Q about things outside your server

Q's knowledge of the wider world has a cutoff date and your question usually does not. So when the answer depends on something that changes over time, Q searches the web mid-answer and continues the same reply with what it found. "Did Discord change anything about bot verification recently" gets a current answer with the pages it came from, not a confident guess from an old memory.

• It shows you the sources. Every web-backed answer carries the links it used. You can open them and read the rule yourself before changing anything on your server.
• It knows what it did not find. If the search does not cover part of what you asked, Q says that part is not covered instead of quietly filling the gap.
• It is not browsing. Q reads search results. It cannot log into anything, fill in a form, or open a page behind a paywall.

Ask Q to write a post

Not everything is a setting. When you want something *posted*, just say so: "tell everyone in #general my stream starts in 5 minutes". Q writes it and hands it back as an editable card inside the chat, with the real Discord preview above it, so you see exactly what your members will see before anyone sees it.

1. Ask Q for the post, naming the channel if you already know it. If you don't, Q asks with a channel picker. When your server makes the answer obvious, that picker opens on the channel Q would use, with one line saying why, so you confirm instead of reading a list. Every other channel is still there under it.
2. Edit the wording right in the card. Q's draft is a starting point, not a decision. You can also just ask for the change in words: say "make the footer shorter" and only the footer changes, because Q reads the card in front of you rather than writing the post again.
3. Need it to look like more than plain text? Open Design for the visual message editor, where you can edit text directly on the preview or switch to fields: title, description, colour, fields, footer, author, and buttons.
4. Add image opens your QAssets library, so anything you have already uploaded is one tap away.
5. Nothing in the library fits? Draw one next to it takes a line of description and gives you options in place. Click one and it goes into the post and into your library. Match the post hands the generator what your draft already says, so the picture is drawn for that post rather than for the words you typed in the box.
6. Press Review message, check the message and button actions, then Confirm and send. Nothing is posted until you confirm.

Not sure how it should look? Ask for a couple of options. Q puts each one on its own tab in the same card, with its own wording and its own design, and you flip between them and see the real preview of each. Edit the one you like, send that one, and the others simply go away.

Visual message editor

When you open Design on a draft card or message settings card in Q chat, the message editor provides a Visual / Fields switch, with Visual active by default.

• Visual mode: click directly on any text in the embed preview (title, description, author name, or footer) to edit it in place. Use the + button on the preview edge to add an image, thumbnail, author, footer, field, or button right where it appears in Discord. Drafts can mix up to five buttons: open a link, give/remove/toggle a role for the member who clicks, or open an existing ticket panel. Pick roles and ticket panels from the server lists.
• Focused panels: adding or editing a field, an image, or a button opens a focused panel (a bottom sheet on a phone, an anchored card on desktop). Soft suggestions offer relevant additions one at a time.
• Fields mode: switches to the structured form layout. Both modes read and write the exact same draft, so switching back and forth never drops or changes content.
• Honest limits and explanations: Discord limits such as 25 fields or total character counts refuse with clear text rather than trimming silently. If AI image generation is unavailable on your plan, the button stays visible as disabled and explains the reason when clicked, pointing to the image library instead.

The same editor sits on the module pages that edit a message: welcome, goodbye and the welcome DM, tickets (closing message), introductions, giveaways, logging, leveling, reminders, reputation, polls and scheduled messages. There it opens in Fields, the form you know, and Visual is one click away. Both edit the same message and Save works as before.

Before sending a draft, Review message shows its destination and every button action and target. Confirm and send publishes that version. Editing clears the review. A refused button leaves your message and other buttons in place so you can correct it.

Change a setting yourself, without leaving the chat

Sometimes you know exactly which knob you want to turn. Say so ("let me change the welcome settings myself") and Q opens that module's settings right there in the conversation: every field at its current value, with proper pickers for channels and roles, not a page you have to go and find.

1. Ask Q to open the settings for the module you have in mind.
2. Change what you want. Everything shows the value your server has right now, so you are editing reality rather than a blank form.
3. Press Save. Only the fields you touched are written, so nothing else gets overwritten.

Send Q a picture of your own

Attach an image to a message in the chat and it lands in your QAssets library straight away, under the filename you sent it with. No upload step, no image host, no pasting a link back. Q can see it and can use it in the same breath.

• It is yours from the moment you send it. The file shows up in QAssets like anything you uploaded there by hand, so you can reuse it later without digging through the chat.
• Ask for it by description. "Put the image I just sent on the reaction roles panel we made" is enough. Q finds the panel and wires the picture in.
• It stays available. Q still has the picture a few turns later, so you can send it first and decide what to do with it after.

Command names are checked against your server

The name of a slash command is the detail an assistant is most likely to get slightly wrong, and a wrong one costs more than it looks: you type it, nothing happens, and you have no way to tell which half of an otherwise correct answer to keep. So after Q writes a reply, Qreate reads your server's real command list from Discord, the same list your client draws when you type a slash, and compares.

If Q named a command your server does not have, a correction appears under the answer saying so and telling you to type a slash and read the real list. That list is yours specifically: the handful Qreate registers everywhere, plus your own modules and any custom commands you built. A correction is never printed for a command that works on your server, and nothing is printed at all if the list could not be read, so no correction is not the same as confirmation.

Posts follow a real layout, not a guess

When Q writes something for you to post, it does not invent a layout. It picks one of a set measured from servers that already run well, and fills it. The draft card shows the layouts it thought were worth offering, so you switch between them and send the one you like.

• Announcement: one thing the whole server needs to know, written as prose with the ping on the line above it.
• One rule: a rules channel is written one rule per post, each with what happens if it is broken in the footer, and one colour across the set.
• Panel: the two lines above a button. The button is the content, so the text only has to get somebody to it.
• Event: the when, the where and the who, each on its own line, so they can be found again rather than read in order.
• Welcome: addressed to the member by name, with their own avatar beside it and the first steps as steps.
• Spotlight: an award or a result, where the colour says which tier it is rather than just looking nice.
• Plain post: no card around it. A quick heads-up does not need one, and Q picking this is an answer, not a failure.

Q is also shown what your server already publishes: the colours you use, how long your posts usually run, whether you use a footer. It matches that where the layout allows it, so a post reads as yours rather than as something generic. If it cannot read any of that, it uses the plain defaults instead of claiming a house style you do not have.

Ask Q to draw the picture

You can ask Q for artwork too: "draw me a server icon, a neon cat in teal and magenta". Q makes three options, at the exact size Discord wants, and you pick one.

• Server icon (512x512): Discord shows it as a circle, so Q keeps the subject centred and the corners empty.
• Server banner (1920x1080): cropped differently on desktop and mobile, so what matters stays in the middle.
• Message image (1280x720): the big picture inside a post Q wrote for you.
• Emoji (128x128), sticker (320x320), role icon (64x64): small targets get simple shapes, because detail turns into a smudge at 20 pixels.

1. Ask for the picture, or accept when Q offers one during a build.
2. Pick the option you like. The others stay archived in QAssets, so you can change your mind later.
3. Save to my library files it in QAssets, named and tagged.
4. If Q also wrote you a post that turn, Use it in the message drops the picture straight into it.

Just talk to it

The waveform button next to the mic opens a spoken conversation. You talk, Q answers out loud, and nobody presses anything to take a turn: Q works out that you have finished a sentence from the pause. Talking over it stops it mid word, the way it works with a person.

Q answers in whatever language you actually spoke, not the language your dashboard happens to be set to. Set your dashboard to English and talk in Russian, and Q listens, replies, and speaks back in Russian, voice included. The first sentence of a session is what settles it, so give it a normal sentence rather than one word. If your browser keeps guessing wrong, pin the language from the selector at the top of the voice screen and it stops guessing.

The screen stays split while you talk. The top half holds whatever Q just prepared, the proposal with its Apply button, a draft message, the artwork options, and it does not scroll away or get covered by the voice controls. The bottom half is the transcript. Q's words light up as it says them so you can always see exactly where it is, and if an answer runs long, the rest waits behind a small "Show the rest" tap instead of filling the screen at once. So you can ask for verification and a welcome flow out loud, hear the one sentence summary, and read the three actual changes before approving any of them.

When there is more than one reasonable way to go, Q offers a couple of buttons to tap instead of listing every option out loud. Nobody can hold seven choices in their head from a sentence.

• The mic button is dictation: what you say lands in the text box and you send it yourself. Best when the exact wording matters.
• The waveform button is the conversation: Q replies out loud and listens again straight away.
• Turn the microphone off with the left button in the dock when someone walks in. The conversation and the cards stay put.
• Type instead hands you back to the keyboard mid conversation, in the same thread.

If an answer takes a moment, Q says so out loud rather than leaving you in silence. You can change which voice it uses from the picker at the top of the voice screen, choosing from eight voices, and the choice is remembered for whichever language you're speaking, on that device, so English and another language can each keep their own.

Q asks before it builds

When you ask for a module, Q repeats back what it understood, asks only the questions it genuinely cannot answer from your server (which channel, which roles, what happens on the first offence versus the fourth), then tells you which defaults it picked for you and what it is not touching. Setting up a brand new server from scratch is a longer conversation on purpose. A one-field change you already spelled out is not: Q just does it.

How much Q can do on its own

Every server has its own answer to "can Q just do this". Open Q's authority (search for it in the dashboard, top bar, or the shield in the chat header) to set it. There are five modes, from least to most: Advisor only explains and never builds anything to approve, Plan writes the whole plan first and waits for your go before building the changes, Ask first proposes changes and waits for your Apply, which is the default and today's behaviour, Trusted sets up and posts routine work on its own while deleting still asks first, and Admin carries out almost everything without asking. Deleting a channel or a role always waits for you, in every mode, and so does acting on one member.

• Six separate capabilities sit under the mode: configuring settings, creating roles and channels, publishing posts, revising what Q posted, deleting a channel or a role, and moderating. You can override any one of them on top of your mode without changing the rest.
• Q can work with anything on the server, not only what it built itself. A channel someone made by hand is fair game the same as one Q scaffolded, so it's the confirmation that keeps you safe, not a check on who created the thing.
• Deleting never runs on its own, at any mode, including Admin. Discord has no restore for a channel or a role, so that one action always shows you the real object and waits for you to press a button, even on the most hands-off setting.

Talking to Q from Discord

You don't have to leave Discord to ask Q for a change. Any admin can run /q right in a server channel: describe what you want in one sentence and Q replies with a plain-language summary and, when there's something concrete to change, an Approve/Cancel proposal card in the same message.

1. Run /q request: followed by your ask, e.g. /q request: turn on the leveling module.
2. Q replies ephemerally (only you can see it) with its answer, and a proposal card if there's a change to make.
3. Click 💬 Continue to keep talking: it opens a small text box and Q's reply updates the same message, so the whole exchange stays in one place.
4. Click Approve to apply a proposed change, or Cancel to drop it and keep chatting.