AI Control
AI Control is the settings page for everything AI on a server. Open it from Configure → AI Control in the dashboard sidebar. It has three tabs: Usage & results, AI Assist (the settings), and Memory (the Memory Library).
For what staff and flows actually do with AI, see AI Assist. This page is the reference for the settings behind it.
What each plan can change
The ticket viewer's AI tools run on every plan, within the server's monthly AI
allowance, and /ask runs on every plan too. Changing the settings below, the
Memory Library, your own provider key, automatic replies, AI flow nodes, and
the /draft, /summarize and /translate commands are Pro. On Free and
Community this page opens as a preview: the settings are visible but disabled,
the defaults apply, and the usage card and results are a sample.
Viewing and changing the page needs the Manage Settings permission (the Admin role). Read-only staff can view it, but the controls are disabled for them.
Usage & results
Results
How AI performed on the server over the last 30 days, from real usage and staff feedback. A value with nothing to go on yet reads -- rather than an invented number.
| Tile | What it measures |
|---|---|
| Deflection rate | How many closed tickets got an AI reply |
| Thumbs-up rate | The share of rated AI replies that were rated up |
| Replies sent | AI replies plus staff suggestions |
| Tokens used | Tokens drawn from the server's budget this month |
On Pro, a Usage breakdown panel splits the period's tokens by type, and each type expands to the flows, automations, or staff behind it.
Tokens and top-ups
Every AI feature on the server meters against one budget: the ticket viewer tools, the Pro Discord commands, AI flow nodes, and the automation AI reply action. The budget is two pools:
- Monthly allowance. The tokens your plan includes each month. It resets with the cycle.
- Top-up tokens. A separate balance that never expires. It is only spent once the monthly allowance runs out.
| AI budget | Free | Community | Pro |
|---|---|---|---|
| AI tokens / month | 10,000 | 250,000 | 5,000,000 |
| Ask AI tokens / month | 0 | 500,000 | 1,000,000 |
Top up on the AI usage card buys more tokens into the top-up balance. It is offered on every plan that includes a monthly allowance, as long as your own provider key is not in use, and needs the Manage Billing permission (Admin or Billing Manager). The live card, with Top up, is on Settings & Access → Plan & Billing on every plan; on Pro it also appears on this tab. A server in an organization draws on the organization's shared pool, so its top-ups are bought on the organization's Plan & Billing page.
On Pro you can also Bring your own key to run AI on your own provider account (not for a server in an organization). Usage is still metered and shown, but the monthly cap no longer applies, and top-ups are not offered while the key is in use.
/ask and the dashboard Ask AI popout have their own Ask AI allowance (the second row above). On paid plans it is spent first; once it runs out, questions draw on the shared budget and top-up tokens, and after that /ask answers from the Ticket Tool docs only. On Free, /ask always answers from the docs only, within daily limits.
The public docs homepage has a separate Ask AI card for product and setup questions. It reads only the published Ticket Tool documentation, cites the pages behind its answer, and cannot see a visitor's server, tickets, members, or private knowledge base. Choose Quick answer for a concise response or Detailed steps for a longer walkthrough. Anonymous requests use CAPTCHA and request limits.
AI Assist settings
The AI Assist tab holds the server-wide defaults for anywhere the bot writes a reply. Changes save with Save changes at the bottom of the card.
Staff tools
| Setting | What it does |
|---|---|
| AI summary | Shows the summarize tool in the ticket viewer. /summarize follows it too |
| AI suggested reply | Shows the quick reply and reply builder tools in the ticket viewer. /draft follows it too |
| AI commands in Discord | Keeps the AI slash commands and the right-click AI actions in your server. Off, the bot removes the Pro commands and /ask answers that AI commands are turned off |
Turn both ticket viewer tools off and the AI button disappears from the ticket header.
Voice
| Setting | What it does |
|---|---|
| Persona | Your own instructions for every AI reply: brand voice, sign-offs, phrases to avoid, house rules |
| Reply style | The baseline tone: concise, friendly, formal, or apologetic |
| Reply language | Auto (the default) answers each customer in the language they wrote in |
| Keep replies brief | Caps automatic replies and staff drafts at about three sentences instead of six |
To pin one language, pick it from the list, or choose Other and type one, such as "Brazilian Portuguese" or "formal Japanese". Set the language here rather than in the persona: the language setting is applied after the persona, so the two can contradict each other.
Grounding
Grounding makes the assistant retrieve your approved Memory entries and knowledge base articles first, then write the answer from them.
| Setting | What it does |
|---|---|
| Ground replies in Knowledge Base + Memory | The server default for automated replies. Reply flow nodes without their own grounding setting inherit it |
| Answer strictness | What happens when nothing relevant is found (see below) |
| Ask for feedback on AI replies | Adds thumbs buttons under AI replies in Discord, for Nobody, Staff only (default), Customer only, or Both |
| Escalation role | The role mentioned when an auto-reply hands a ticket to a person |
| Use knowledge base | Lets AI read your knowledge base articles for context |
| Use the customer's past tickets | Lets AI read the customer's previous tickets for context |
| Strictness | Behavior |
|---|---|
| Strict | Replies only when approved knowledge matches. Otherwise it stays silent and escalates |
| Balanced | The default. Prefers relevant knowledge and falls back to a general answer |
| Lenient | Always attempts an answer, even with no match |
Strict needs something to match
Until the server has at least one published knowledge base article or one approved Memory entry (including entries your organization shares), strict behaves like balanced, because an empty library could only ever refuse. The setting still reads Strict, and strict answers begin once the first article or entry exists.
Auto-reply limits
| Setting | What it does |
|---|---|
| Reply to first customer message | Answers a customer's first message in a ticket with a grounded reply, with no flow to build |
| Max auto-replies per ticket | Caps first-message and automation AI replies in one ticket, so a back-and-forth never becomes a loop. 0 means no cap |
| Auto-reply categories | Limits Reply to first customer message to these categories. Empty means all |
For a reply a person approves first, use an AI Auto Reply node in draft mode instead of the toggle. See AI Assist.
Memory capture
| Setting | What it does |
|---|---|
| Learn from resolved tickets | When a ticket closes, distills its question and resolving answer into a memory candidate |
| Learn from thumbs-up and edited replies | A thumbs-up on an AI reply, or a draft staff edited before sending, becomes a candidate |
| Auto-approve answers that keep coming up | On by default. Approves an answer once three resolved tickets give the same one. A staff thumbs-down on a reply that used it sends it back. Candidates nobody reviews are cleared after 30 days |
| Memory capture categories | Limits capture to these categories. Empty means all |
Staff can also save a message straight into the library as an approved entry. In Discord, right-click the message, open Apps, and pick Save to AI Memory. In the dashboard ticket view, right-click a message and pick Save as AI fact or Save as AI Q&A. Saving needs Pro. In Discord any support staff member can save; in the dashboard it takes permission to manage the knowledge base.
Memory Library
The Memory tab is where you curate what the assistant may reuse. Only approved entries ground a reply, whether staff approved them or auto-approve did. Candidates wait for review, and rejected entries are kept for the record but never reused.
Each entry is one of two kinds, chosen when it is saved:
| Kind | What it holds |
|---|---|
| Question & answer | A customer question and the answer that resolved it. Best for recurring questions |
| Fact | One statement that stands on its own. Best for policies, prices, limits, anything always true |
From the library you can:
- Approve, edit, tag, reject or delete entries, one at a time or in bulk.
- Filter by status, by kind, by source (manual or from tickets), by tag, or to entries that have never grounded a reply.
- Export the library as JSON or CSV, and import entries the same way. Imported entries are approved immediately, and rows with no question become facts.
- Add an entry by hand. The library warns you when a near-duplicate already exists.
Servers in an organization also see entries the organization shares. They are managed on the organization page, are retrieved in every member server, and show read-only with an org badge.
Your answers, not model training
Memory is scoped to your server, or to your organization for shared entries. None of it is used to train an AI model: the assistant only retrieves approved answers, ones your staff approved or ones auto-approve promoted because they kept coming up. Turn auto-approve off to review every answer yourself.
Related
- AI Assist: what staff and flows do with AI
- Automations: the first-message trigger and the AI reply action
- Knowledge Base: the articles AI answers ground in
- How Billing Works: plans and AI token budgets