AI Features Overview

What TicketCord's AI features do, where to turn them on, which plan each one needs, and how tokens work

TicketCord's AI, Atlas, answers questions from your own help content, sorts and prioritizes tickets, translates conversations, and warns your team when a member is upset. Each job has its own switch per server, so you can start with one and add more when you are ready. Everything else is set up for you with defaults that work for most servers. This page is for the server admin who sets them up.

AI features require a Pro or Enterprise plan on the bot owner's account. Move tickets to the right category (sorting), Translate tickets, the Translation limit, and Mood history are Enterprise only. Atlas answers by email in the email inbox and in the website widget on Pro and Enterprise. The 14-day signup trial runs as Enterprise with a one-time allowance of 1,000,000 tokens. View pricing

Available AI features

The What Atlas does tab also holds four smaller jobs, each with a switch on its card:

  • Ticket summary: a short note for staff on each new ticket, on the ticket page and in /claims: what the member needs, how they feel, and how urgent it is. When an email follows up on a closed email ticket, the new ticket also gets a summary of the earlier one (see More emails from the same person).
  • Ask for details when a member only says hello: when a ticket starts with only a greeting, Atlas asks the member what they need, or pings a role. In a ModMail ticket the question goes to the member's DMs. Change its wording on the Messages page (Ask for details).
  • AI Help drafts: an AI Help button on the ticket welcome message and the /ai-help command, which draft a reply for staff to review, in each staff member's own writing style when they set one with /ai-style, and the AI Help panel on the dashboard's ticket page, which also answers staff questions about the ticket. See AI Help for staff.
  • Remember members: Atlas learns short support facts about each member from their closed tickets and uses them in their later tickets and in AI Help drafts. On by default. See Member Memory.

All four are Pro and higher.

Where the settings live

Go to AI assistant → Atlas

Open the server

Go to your server, or click its tile on the dashboard home.

Open AI assistant

Go to AI assistant, under Settings in the sidebar. The page has five tabs:

  • What Atlas does: the Atlas switch, AI tokens, AI usage, and every job, in sections: Answer members, Help your staff, Sort and flag new tickets, Languages, and Limits and who Atlas skips.
  • Help content: what Atlas answers from: Sources, About your server, Where else Atlas may look, and Learning.
  • Test: for the bot owner, ask Atlas a question and see what it would do in a ticket, or replay your latest closed tickets next to your staff's replies, with nothing posted. See Test Atlas.
  • Staff Q&A: the question-and-answer entries Atlas learned or you added, ready to review. A count on the tab shows learned answers waiting for your review. Under the list, Unanswered questions shows what members asked that Atlas could not answer, so you can add the answers.
  • Stats: how your sources and answers perform.

Search for a job by name with Search settings to choose a server and jump straight to its card.

Turn on the Atlas switch

Switch on the Atlas card at the top of What Atlas does. It is one switch for all AI on the server. While it is off, the page says Atlas is off and no job runs, but you can still turn jobs on and change them; they start when you turn Atlas on. Your choices are kept when you turn it off.

Turn on the jobs you want

Switch on each job on its card. A line under the Atlas card says how many jobs Atlas is doing. On Pro, Mood history, Move tickets to the right category, Translate tickets, and Translation limit show locked cards with an Enterprise badge.

Save

Your changes collect in a bar at the bottom of the page, across both settings tabs. Click Save changes to apply them or Undo changes to drop them. Nothing is applied until you save.

Server-wide settings

These settings apply to every AI job on the server. Atlas is at the top of AI assistant → Atlas; the rest are in Limits and who Atlas skips at the bottom of What Atlas does.

SettingOptionsDefaultWhat it does
AtlasOn, OffOffOne switch for all AI on this server: every job, the staff commands /ai-help, /summary, and /translation, and Use examples and Preview on the /ai-style card.
Limit per memberOn, Off; Tokens per member (1,000 to 100,000,000); Resets every Day, Week, Month, or YearOff; 100,000 tokens a monthStops one member from using up your tokens. It covers every AI job in that member's tickets and help threads. Staff are never limited, and AI Help drafts staff ask for do not count toward the member's limit. When a member reaches the limit they are told, and Atlas does nothing for them until the next period. The bot owner can clear usage with Reset usage: Reset one member… or Reset everyone….
Keep Atlas away fromUp to 500 membersEmptyAtlas does nothing in tickets these members open: no answers, AI Help drafts, summaries, priority, tone alerts, duplicate checks, sorting, translation, mood history, Ask for details, close requests, learning, or member memory. Staff still see and answer them. Manage opens the list, where you add members by name or Discord user ID. The card also says how many panel buttons turn Atlas off for the tickets they open.
Translation limit (Enterprise)5,000 to 1,000,000 characters a month10,000How many characters Atlas translates each month. See Translation.
Go to AI assistant → Atlas → Keep Atlas away from

AI model and tokens

Atlas runs every job on every server on the same AI model. There is nothing to pick: the Lite and Premium modes that earlier versions offered are gone, and a server that had chosen either now runs on the same model as every other server.

  • Every AI response uses tokens from the bot owner's monthly allowance, billed at 1.15 tokens for each token the model uses.
  • The AI tokens card at the top of AI assistant → Atlas shows the bot owner how many of this month's tokens are left, with a Buy more link, and warns when most of them are used. Dashboard staff see the card without the balance.
  • When the tokens run out, Atlas stops answering and the other AI features pause until the allowance refills or the owner buys extra tokens. TicketCord emails the bot owner the first time this happens each month; on a trial, the bot tells the owner in Discord instead.
  • With extra tokens on the account, AI keeps running on them once the monthly allowance is used up. TicketCord emails the bot owner when that happens, at most once a month; on a trial there is no email.
  • Atlas needs a Pro or Enterprise plan on the bot owner's account.

Atlas runs on an AI model we choose. When a better model becomes available, we move to it without announcing the change. Before any switch, we test the new model on our own support evaluations and only switch when it does at least as well. Token use per response can change with the model, so AI usage is the place to see real consumption.

Go to AI assistant → Atlas → AI tokens

AI usage

The AI assistant → Atlas → AI usage card, under AI tokens on What Atlas does, sets how much of Atlas's work runs on one server. It is a per-server setting: each server of a bot has its own choice. The card shows while the Atlas switch is on. Only the bot owner can change it; dashboard staff see the choice. It is not the AI usage page, which shows how many tokens the account used.

ChoiceWhat Atlas does
DefaultAtlas runs everything that is switched on. Recommended, and what every server uses until the owner changes it.
Essentials onlyUses fewer AI tokens. Atlas still answers questions, and AI Help still works when staff ask for it. Its background work is paused.

While Essentials only is selected, these are paused, whatever their own switch says:

  • Ticket summary updates after the first one. Each ticket still gets its summary once. It is no longer updated as the member writes more. Staff can still update it with Update on the summary on the ticket page.
  • Mood history. Tickets that close are not read for mood.
  • Ticket reviews. Tickets that close are not reviewed.
  • Remember members. No new facts are learned from tickets that close. Facts already kept are still used.
  • Learning from closed tickets. No new answers are learned from solved tickets.

Tone alerts, priority, duplicate checks, sorting, and translation are not affected.

The trade-off: answers can become less accurate over time, because Atlas stops learning from closed tickets and stops remembering members.

Nothing is switched off for you. Each paused job keeps its switch and its settings, its card says that it is paused, and choosing Default again and saving brings every one of them back as it was. Essentials only needs a Pro or Enterprise plan on the bot owner's account; Default can always be chosen.

Go to AI assistant → Atlas → AI usage

What happens

Every job follows the same pattern:

  1. Something happens in a ticket: it is created, or the member sends a message.
  2. Atlas waits a moment so it reads follow-up messages together. Checks that need the whole problem (priority, tone, sorting, and the ticket summary) wait until the member has explained it, at most about 30 seconds. Ask for details waits up to 30 seconds and stays quiet if the member writes again. Answers never talk over staff who are typing.
  3. Atlas reads the conversation and decides what to do. When it is not sure enough, it does nothing rather than guess.
  4. The action runs: an answer is posted, the ticket is moved, the channel is renamed, or an alert is posted for staff.
  5. Where a job has feedback (members rating answers, learning from solved tickets), your team's input improves later answers.

Alerts for staff (upset members, priority changes, duplicate tickets, sorting, and what Atlas learns) post in the AI alerts channel set under General → Where the bot posts. When no AI alerts channel is chosen, they post in your logs channel. When neither is set, alerts are not posted, but the jobs still run. The cards on the AI page that post alerts say where they go.

AI tokens

Tokens are the budget every AI job draws from. Each plan includes a monthly allowance:

PlanTokens per month
Free0 (no AI access)
Basic0 (no AI access)
Pro10,000,000
Enterprise30,000,000
  • Usage refills every month, on annual plans too. Open AI usage (in the sidebar's More tools group on the dashboard home, or in the account menu) to see tokens used, when they refill, the daily average, and how many you are expected to use by the reset.
  • Need more? Token packages of 10M, 20M, or 50M tokens are available under Extra tokens on the AI usage page, and Enterprise triples the monthly allowance.
  • How many tokens a job uses depends on how long the messages are; every response is billed at 1.15 tokens per token the model uses. The AI usage page is the reliable place to see real consumption.
  • Reading knowledge sources uses tokens too: when a web page, GitHub or file source is read or read again, AI prepares its content so Atlas can answer from it. Under By feature, the AI usage page lists this separately as Reading knowledge sources, apart from Knowledge base, which is what answering questions used. Days before this was listed separately show it under Knowledge base. See Approving the AI cost.
  • To use fewer tokens on a server without turning jobs off one by one, choose Essentials only under AI usage.
  • On the signup trial, AI features pause when the 1,000,000-token trial allowance is used up. Upgrading restores the full monthly allowance.

Limits and gotchas

The plan check uses the bot owner's plan, not the plan of the dashboard staff member changing settings. If the owner downgrades, AI features stop even though the switches still look on.

  • Keep Atlas away from, and a panel button set to turn Atlas off, stop every AI job in the tickets they cover, including AI Help drafts and /summary, which then answer with TC018.
  • Outside hours under When staff are around on AI assistant → Atlas → Answer questions in tickets needs business hours, which are an Enterprise feature, and is offered only while business hours are on.
  • When Atlas cannot help a member, it always offers to hand the ticket to staff and then stays out of it. This cannot be turned off; you choose how long Atlas stays out. See Feedback & Learning.

Troubleshooting

SymptomCauseFix
The page says Atlas is off and nothing happens in ticketsThe Atlas switch is offSwitch on the Atlas card at the top and save
The page says Atlas is on, but no jobs are turned on yetEvery job's switch is offTurn on the jobs you want on their cards and save
A card shows a Pro+ or Enterprise badge and its switch is greyed outThe bot owner's plan is below the feature's requirementUpgrade the owner's plan (the badge links to Plan & billing); dashboard staff cannot unlock it
Jobs are on but nothing happensThe owner's monthly tokens are used up, or the trial allowance is exhaustedCheck AI usage; wait for the reset or add a token package
Atlas does nothing for one member but works for othersThey reached their Limit per member, are on Keep Atlas away from, or opened the ticket from a panel button that turns Atlas offReset them with Reset usage on AI assistant → Atlas → Limit per member, remove them from the list, or change the panel button
Alerts never appearNo AI alerts channel and no logs channel is set, or the bot cannot post thereChoose a channel under General → Where the bot posts (AI alerts, or the logs channel) and give the bot View Channel and Send Messages in it
/ai-help, /summary, or /translation replies with TC018Atlas or that job is off, or the ticket is one Atlas is kept away fromSee TC018
The bot replies with TC014The action needs a higher planSee TC014

Common questions

Do I need to turn on every AI feature? No. Each job has its own switch. A common start is Answer questions in tickets alone, then Set priority automatically once you trust the answers.

Can I try Atlas before members see it? Yes. The Test tab of AI assistant → Atlas → Test Atlas shows what Atlas would answer to a question, with its sources, without posting anything. It works while answers are still off. See Test Atlas.

Do I need to tune any settings? No. TicketCord sets timings, thresholds, and quality checks for you. Each card only asks for the choices that differ between servers, such as which role to ping or how long Atlas stays out of a ticket after handing it to staff.

Do I need to pick an AI mode? No. Atlas runs every server on the same model, and the old Lite and Premium choice is gone. Each response is billed at 1.15 tokens per token the model uses; the AI tokens card at the top of AI assistant → Atlas shows the bot owner what is left this month.

Which AI model does Atlas use? Atlas runs on a model we choose, and we move to a better model when one becomes available, without announcing the change. We test every new model on our own support evaluations first and only switch when it does at least as well.

Can dashboard staff change AI settings? Yes, if they have dashboard access, except for owner-only actions: choosing AI usage (Default or Essentials only), managing text file and GitHub repository sources, status page integrations, and resetting member token usage.

Do AI features work on the Free or Basic plan? No. Both plans have no AI access and 0 tokens. Pro and Enterprise include AI. The signup trial gives Enterprise access for 14 days.

Which features are Enterprise only? Move tickets to the right category (sorting), Translate tickets, the Translation limit, and Mood history (the mood journey on the ticket page). Translate ModMail is available from Pro.

What happens when tokens run out? AI features pause until your monthly tokens refill or until you add a token package. Tickets keep working normally without AI, and a ticket summary already on a ticket stays there. Open AI usage to see how many tokens are left.

Does the AI read every message? Only messages in tickets and help channels on servers where a job is switched on, and never in tickets from members on Keep Atlas away from. Staff messages are used to decide when Atlas should stay quiet and, for learning, which answers were correct. With Remember members on, Atlas also reads what the member wrote when their ticket closes, to keep short facts about their setup (Member Memory).

Ticket Reviews

Ticket reviews is an optional Pro and Enterprise feature, off by default. In AI assistant → Atlas → Ticket reviews, turn it on and choose the percentage of eligible closed tickets to review. Set Tickets to review (%) to 100 to review every eligible ticket, or use a smaller sample on busy servers. Atlas must be enabled, but automatic answers and mood history can remain off. Reviews use the bot owner's AI tokens. While AI usage is set to Essentials only, tickets that close are not reviewed.

Go to AI assistant → Atlas → Ticket reviews

After a ticket closes, each staff member who replied receives a separate review of professionalism, tone, accuracy, resolution and policy adherence. Reviews describe how their tone changed, identify rude or dismissive replies and cite message IDs. Accuracy needs relevant knowledge-base evidence; policy adherence needs explicit server instructions. Missing evidence is shown as unknown and excluded from the average. Response timing is context, not a score: an overnight wait or customer silence does not establish poor staff performance.

Read the reviews in Stats → Your team or on an individual staff member's page. Owners can see all staff reviews; staff with dashboard access can see their own reviews. These are AI assessments to check against the conversation, not employment decisions.

AI-only tickets, excluded members and tickets where Atlas is muted are not reviewed. Reviews apply to newly closed tickets, not your historical backlog. Reopening and closing a ticket replaces its previous review. Deleted content is not reviewed; existing review text is removed with ticket content. Reviews can be delayed or fail when tokens or services are unavailable. Owners can see waiting and failed counts. Conversations above 1,000 text messages or 160,000 bytes of text cannot currently be reviewed; no partial score is presented as a full-ticket review.

Last updated on

On this page