AI Assist
AI Assist is the set of AI capabilities that help your team answer faster and keep answers accurate. It starts in the ticket viewer, where staff can summarize a conversation and draft a reply, and extends into your flows and automations, where the bot can answer from your knowledge base, reply grounded in real facts, and recall how similar tickets were resolved before.
Pro feature
The AI features on this page are part of the Pro plan. On Free and Community, the AI button and the AI Control page show a preview so you can see what they do, and opening them explains how to upgrade.
The AI button in the ticket viewer
Open a ticket on the dashboard and you will find an AI button in the action bar, next to Claim and Close. The button stays in the action bar at every screen size. The panel has two tabs: Summary and Suggested reply.
Both tabs are grounded in the same approved knowledge the Discord bot answers from: your knowledge base articles and your approved Memory entries. A summary or draft in the dashboard and an auto-reply in Discord draw on one shared answer set, so what staff see and what customers get never drift apart. Your server's persona from AI Control applies to drafts too, so suggestions arrive already in your voice.
Summary
The Summary tab gives you a quick read on the ticket so you do not have to scroll the whole thread.
Conversation summary
A short recap of what the ticket is about and where it stands, drawn from the messages so far.
Sentiment
A badge showing whether the user reads as positive, neutral, or negative, so you can spot the tense ones at a glance.
The summary loads the first time you open the panel. Use Regenerate to refresh it if the conversation has moved on.
Suggested reply
The Suggested reply tab drafts a response you can shape before it goes out. The draft is grounded in your knowledge base and approved Memory, and follows your persona.
Set the tone
Pick a style: concise, friendly, formal, or apologetic. The draft follows it, on top of your server persona.
Steer it (optional)
Add a short instruction, for example "ask for their order number", and regenerate. The draft takes your steer into account.
Edit
The draft lands in an editable box. Tidy it up however you like before sending.
Send or copy
Choose how it goes out:
- Send as self posts the reply under your name
- Send as bot posts it as the bot, when anonymous replies are enabled
- Copy to message box drops the text into the composer so you can keep editing
Canned replies, for the non-AI moments
The reply composer also has a canned replies button for plain saved snippets: your team writes them once on the server's Canned Replies page, and any staff member inserts one with a click. No tokens, no generation, just the exact words you saved, and it is free on every plan. Unlike the AI tools on this page, nothing here needs Pro.
AI commands in Discord
The same tools exist as slash commands, so staff can work a ticket without leaving Discord.
/ask is registered on every plan. On Free it answers product questions from the Ticket Tool documentation. On Community and Pro it also answers questions about your own server, such as ticket counts, stats, and settings, drawing on your knowledge base and approved Memory. The answer is shown only to you.
The remaining commands below are Pro only: the bot registers them in your server on the Pro plan, and on other plans they do not appear at all.
/summarizeposts an AI summary of the current ticket. It follows the AI summary toggle./draft [guidance]drafts a suggested reply, shown only to you, with Send, Edit, and Discard buttons. A Send as brand button appears when anonymous staff replies are enabled. It follows the AI suggested reply toggle.
There is also a message right-click action: right-click a customer message, open Apps, and pick Reply with AI. The bot drafts a grounded reply that addresses that specific message, with the same send controls as /draft.
The full command list, including /canned and /translate, is in the command reference. The AI commands in Discord toggle on AI Control turns the whole set off for your server.
Answers from your knowledge base
The AI KB Answer flow node answers a customer's question from your knowledge base. It retrieves by meaning, not just matching keywords, so a customer who asks "my card was declined" finds the article about failed payments even when those exact words never appear in it.
The node answers only from the articles it retrieves. If your knowledge base does not cover the question, it does not guess. Instead it takes the escalate branch, so you can route the ticket to staff. That keeps automated answers honest: a real answer when there is one, a handoff when there is not.
Meaning plus keywords
Retrieval blends meaning-based search with keyword search, so it catches both the intent of a question and exact things like an order ID, SKU, or product name that a phrasing-only match would miss.
Grounded replies
AI replies can be grounded in your knowledge base and approved memory. When grounding is on, the bot pulls the most relevant facts for the latest customer message and writes the answer from them, instead of guessing from the conversation alone.
Grounding has a server-wide default on the AI Control page, and individual AI Auto Reply and AI Suggested Reply nodes inherit it unless you set Include knowledge base per node. AI Control also sets the answer strictness: in strict mode, an auto-reply that finds nothing to ground in stays silent and can ping an escalation role instead of guessing. Strict needs something to match against, so a server with no published articles and no approved memory is held at the balanced setting until it has one, then strict starts applying on its own. AI Control covers strictness in full.
| Reply type | AI Suggested Reply | AI Auto Reply |
|---|---|---|
Who sees it first | Staff only, as a draft | Can post to the customer |
Grounded in your knowledge | Yes, when enabled | Yes, when enabled |
Best for | A human sends the final word | Fast answers to common questions |
A Suggested Reply is always reviewed by a person: it drafts text for staff and never posts on its own. An Auto Reply can post directly, or run in draft mode (below) so a human approves it first.
Reply to a customer's first message
You can have the bot answer the moment a customer opens a ticket, with no flow to build.
One-click toggle
Turn on Reply to first customer message in AI Control. On a customer's first message the bot writes a grounded reply and posts it.
Or build a rule
Prefer full control? Use the automation trigger On customer's first message and pick your own action, in draft or send mode.
The toggle lives on the AI Control page, along with the dials that bound it: which categories it covers and how many automated replies a single ticket may receive. The trigger is documented in Automations. Either way the first reply is grounded in your approved knowledge, so the customer often gets a useful answer before a staff member picks the ticket up.
Draft or send: review mode
An AI Auto Reply can run in one of two modes, so you decide how much trust to hand it.
Send mode
The reply posts straight to the ticket. Best once you trust the answers for a category of common questions.
Draft mode
The reply is posted as a private, staff-only suggestion with Send, Edit, and Discard buttons. Nothing reaches the customer until a staff member sends it.
Draft mode is the control for turning auto-reply on with confidence. Only support staff can act on a draft: the buttons check permissions, so a customer can never send or edit one. Start in draft mode, watch the suggestions for a while, then switch to send mode when you are happy.
Handing a ticket to a person
Two things take a ticket away from the AI and give it to your team.
The customer asks
Someone who asks to speak to a person gets one short reply saying a staff member is being brought in, in whatever language they wrote in, instead of another attempt at an answer.
Nothing to answer from
When the question is not covered by your approved knowledge, the AI hands over rather than guessing at it.
Either way the bot posts a note in the ticket mentioning your escalation role, then stops replying automatically in that ticket, so it is not talking over the staff member who picks it up. This only stands down the automatic replies: staff can still use /ask, /draft and the AI panel on the ticket as normal.
Set an escalation role
The Escalation role on AI Control is what makes the handoff visible. Without one the AI still stands down when a customer asks for a person, but nobody is mentioned, so the ticket can sit waiting with no one told. The same applies if the role you picked was since deleted.
An AI Auto Reply running in draft mode is exempt from all of this. A staff member is already reading every suggestion before it goes out, so there is nothing to stand down from.
Memory: recall past resolutions
Beyond your hand-written articles, the assistant can recall from past resolutions: how a similar question was answered and closed before. That knowledge grounds future replies the same way an article does.
Per-server and staff-approved
Memory is scoped to your server and is never shared across servers (except entries your own organization chooses to share), and none of it is ever used to train an AI model. The assistant only reuses answers you approved. Nothing a customer said is reused in a reply until a staff member has reviewed and approved it.
There are three ways an answer gets into memory:
A ticket is resolved
When a ticket closes, the system distills the question and the resolving answer into a candidate memory entry. The capture policy in AI Control sets whether this runs and for which categories.
Staff save a message
In Discord, right-click any message in a ticket channel, open Apps, and pick Save to AI Memory. Because a staff member chose it deliberately, it is saved as approved and can ground replies right away.
Feedback becomes a candidate
A thumbs-up on an AI reply, or a draft a staff member edited before sending, becomes a candidate too, when feedback capture is on. The answers your team endorses in the course of normal work flow into the review queue on their own.
Candidates wait in the Memory Library on the AI Control page, where an admin can approve, edit, tag, or reject them, one at a time or in bulk. Only approved entries are ever retrieved. Rejected ones are kept for the record but never reused.
Feedback that builds your library
Small signals from your team now feed the library directly, at no extra token cost. This never trains or fine-tunes an AI model: it only grows and shapes the set of your own approved answers the assistant retrieves from.
Thumbs up or down
Staff can rate an AI reply or suggestion. The votes feed the results shown in AI Control, and a thumbs-up sends that answer to the review queue as a memory candidate.
Edits are corrections
When a staff member edits a draft before sending, the edited version is captured as a candidate, so the better wording is what your team reviews and approves.
The loop is simple: your team answers tickets, endorses the good replies, and an admin approves the keepers. The more answers you approve, the more the assistant can ground a reply in.
Your models, your data, and cost
Bring your own key
Use your own AI provider key to run replies on your account. Token usage is still metered and shown, and bringing your own key removes the monthly cap.
Metered and visible
Every AI request draws from your server's monthly token budget, shown on the AI Control page. Meaning-based search runs on a separate track from the chat replies.
It can always hand off to a human
AI answers only from what it can ground in. It hands the ticket to a person when it cannot answer with confidence, and whenever a customer asks to speak to one. It drafts and suggests. It never silently makes destructive changes, and a staff member is always in the loop for anything it is not sure about.
Budgets and limits
Every AI request draws from your server's monthly AI token budget, the same pool the AI flow nodes and AI Assist share. If the budget runs out, the panel says so plainly and asks you to upgrade or wait for the next cycle. Bringing your own API key removes the cap. See the live per-server token budgets on the AI Control page and in How Billing Works.
Related
- AI Control: the per-server hub for usage, voice, grounding, and the Memory Library
- Managing Tickets: the ticket viewer in full
- Automations: the first-message trigger and AI reply action
- Using Support Flows: AI nodes for automated replies
- How Billing Works: plans and AI token budgets