Skip to main content

Documentation Index

Fetch the complete documentation index at: https://celestia.mintlify.app/llms.txt

Use this file to discover all available pages before exploring further.

Overview

Seraphine is an AI-powered moderation bot that works alongside Celestia to automatically monitor messages in your server. It uses large language models to detect clear rule violations in real time and reports them directly to Celestia for action. Seraphine is lenient by design — it only flags obvious violations and ignores casual language, banter, and edgy humor that is normal in Discord communities. Seraphine requires a one-time unlock purchased through Celestia and can only operate in servers where Celestia is present.

Requirements

  • Celestia must be present in your server
  • The user who adds Seraphine must have the Seraphine unlock purchased through Celestia’s store
  • AI moderation must be enabled per server using /seraphine enable
If Celestia is removed from your server, Seraphine will automatically leave as well.

Commands

Enables AI moderation for your server. Once enabled, Seraphine will begin scanning messages and reporting violations to Celestia.
/seraphine enable
Permissions: Server Owner only
Disables AI moderation for your server. Seraphine will stop scanning messages entirely until re-enabled.
/seraphine disable
Permissions: Server Owner only
Opens a modal where you can write custom rules for your server. Seraphine will use these rules instead of the default ones when analyzing messages.
/seraphine rules set
Permissions: Server Owner only
Write rules in a numbered format for best results. Keep each rule short and specific. Vague rules like “be respectful” may produce inconsistent results.
Displays the current rules Seraphine is using for your server. Shows custom rules if set, otherwise shows the default rules.
/seraphine rules view
Permissions: Server Owner only
Clears your custom rules and resets Seraphine back to the default rule set.
/seraphine rules reset
Permissions: Server Owner only
Whitelist a user, role, or channel from AI moderation. Whitelisted targets are completely ignored by Seraphine.
/whitelist add [user] [role] [channel]
Permissions: Manage Server
Remove a user, role, or channel from the whitelist.
/whitelist remove [user] [role] [channel]
Permissions: Manage Server
View all currently whitelisted users, roles, and channels for your server.
/whitelist list
Permissions: Manage Server
View all AI warns for a user in your server, including the rule broken, severity, date, and the message content that triggered the violation.
/warns list <user>
Permissions: Manage Server
Remove a specific warn from a user by its number. Use /warns list first to find the warn number.
/warns remove <user> <index>
Permissions: Manage Server
Clear all AI warns from a user in your server.
/warns clear <user>
Permissions: Manage Server

Default Rules

If no custom rules are set, Seraphine uses the following default rules:
  1. No racist remarks, slurs, or hate speech targeting someone’s race, ethnicity, or nationality
  2. No NSFW content, sexual discussions, or explicit material in non-designated channels
  3. No targeted harassment or threats against members

What Seraphine Flags

Seraphine is intentionally lenient. It is built for obvious, clear violations only.

Flagged

  • Actual racial slurs or racist remarks directed at a person or group
  • Explicitly sexual messages or graphic descriptions
  • Direct threats or serious harassment

Not Flagged

  • General swearing or cursing
  • Friendly trash talk or banter
  • Dark humor not targeting a race or being explicitly sexual
  • Mild insults or name calling between friends
  • Casual phrases like “i’ll kill you” or “you’re stupid”
  • Any ambiguous message
If Seraphine is unsure whether a message is a violation, it will always return clean. False negatives are preferred over false positives.

Violation Severity

When a violation is detected, Seraphine assigns a severity level and reports it to Celestia:
SeverityDescription
LowBorderline case, mild violation
MediumClear and obvious violation
HighSevere slur, explicit content, or direct threat

How It Works

  1. A message is sent in your server
  2. Seraphine checks if AI moderation is enabled for the guild — if disabled, nothing happens
  3. Checks if the user, role, or channel is whitelisted — if so, skips the message
  4. Sends the message to an AI model along with your server’s rules
  5. If a violation is detected, reports it to Celestia with the user, channel, message content, rule broken, and severity
  6. Celestia handles the warning and any further action

Custom Rules

You can replace the default rules with rules specific to your server using /seraphine rules set. This opens a text modal where you can write up to 1000 characters of rules.
Number your rules and keep them specific. Example:
1. No promoting or discussing illegal activities
2. No hate speech of any kind
3. No sharing personal information of other members
You can view them anytime with /seraphine rules view and reset them with /seraphine rules reset.

Whitelist

The whitelist allows you to exclude specific users, roles, or channels from AI moderation entirely. This is useful for staff channels, trusted roles, or specific users who should not be scanned.
Always whitelist your moderation staff and trusted roles to avoid false reports.

Warns

Every violation detected by Seraphine is automatically saved as an AI warn against the user in your server. Warns are stored per user per server and include a full history of each incident. Each warn entry contains:
  • The rule that was violated
  • The severity level
  • The message content that triggered the violation
  • The channel and message it came from
  • The timestamp
You can view, remove, or clear warns at any time using the /warns commands. Warns are independent per server — a user warned in one server is not affected in another.

Logging

When Seraphine detects a violation, it reports it to Celestia which handles the action in the same server. No separate log channel configuration is required.