Make it yours
Guardrails and backups
Guardrails are the limits you set for the AI. Find them in /settings > Guardrails, or see them all with /config view guardrails. /config explain describes any setting in plain English.
What you can set
- Blocked words: never allowed in questions, picture requests or letters.
- Blocked topics: subjects the AI politely refuses, such as politics or cheats.
- Always call a human: if a ticket mentions any of these words, staff are called straight away. Default: refund, ban appeal, harass, threat, chargeback.
- AI disclaimer: a short line under every AI answer, such as AI answer, staff have the final say.
- Real people in pictures: off by default. We strongly recommend leaving it off.
- Safety filter: runs picture requests and letters through a free safety check (on by default).
- Theme check for pictures: checks every picture request fits your theme and rules first (on by default). On a themed server, harmless off-theme requests are adapted rather than refused.
- Longest picture request: 400 characters by default.
For spending, each feature has its own limits: pictures per month, per member per day and the wait between them; help desk questions per member per hour; letters per member per day. Every feature also has its own on/off switch.
Safe limits
You can tune every setting, but only inside safe bounds, so the AI can't be left without brakes (no unlimited retries, no zero cooldowns, no huge prompts). A value outside its range is refused with an explanation, and an old backup with one is brought back inside it. /config explain shows each setting's range.
| Setting | Lowest | Highest |
|---|---|---|
guardrails.max_prompt_length | 20 | 1000 |
assistant.questions_per_user_per_hour | 1 | 60 |
assistant.max_answer_words | 30 | 400 |
tickets.escalate_after_attempts | 1 | 8 |
tickets.max_open_per_user | 1 | 5 |
images.per_user_daily | 0 | 50 |
images.cooldown_seconds | 10 | 86400 |
images.monthly_cap | 0 | 100000 |
letters.per_user_daily | 0 | 50 |
letters.max_length | 50 | 4000 |
insights.min_length | 5 | 2000 |
feedback.min_length | 5 | 500 |
watch.min_confidence | 0.5 | 0.99 |
watch.strike_expiry_days | 0 | 3650 |
polls.default_hours | 1 | 768 |
economy.message_cooldown_seconds | 5 | 86400 |
moderation.timeout_minutes | 1 | 40320 |
moderation.spam_window_seconds | 2 | 600 |
moderation.spam_max_messages | 2 | 100 |
Text the AI reads with every request is kept short too: the server description up to 1,500 characters, the AI voice up to 300, up to 30 AI watch rules of 200 characters each, up to 40 blocked topics, and up to 10 auto-answer channels.
On top of that, each plan has a daily AI cap that can't be switched off, and a best picture quality:
| Plan | AI answers, ticket replies and letter rewrites a day | Best image quality |
|---|---|---|
| Free | 60 | No images |
| Standard | 600 | high |
| Plus | 2,000 | xhigh |
| Premium | 6,000 | max |
The help desk's only answer from knowledge switch (on by default) makes it say it doesn't know rather than make things up.
Tell members how AI is used
In your rules or info channel, run /settings, pick Guardrails and press Post an AI notice here. Journo posts a plain notice listing where its AI reads and answers: the help desk, AI ticket replies, the channels the AI watch checks and the channels member insights read, with how long summaries are kept. Press it again after you change channels.
Backups
/config exportdownloads all your settings as a file./config importloads them back, or copies them to another server.