Question 1

What happens to my server if the decision API goes down mid-conversation?

Accepted Answer

It fails open. A connection error, invalid key, rate limit or unexpected exception lets the message through unmoderated rather than deleting it. Three consecutive failures in a server post one outage alert to the configured mod-log channel, further failures are deduplicated rather than spamming it, and the next successful evaluation posts a recovery notice.

Question 2

Will it delete legitimate messages before any human sees them?

Accepted Answer

That risk is designed around rather than ignored. A flagged message is deleted immediately with the reason written to the Discord audit log, and the mod-log card carries one-click pardon and ban buttons. When an admin pardons, the text is stored and fed into future prompts for that server as a verified legitimate example, so it picks up local slang without retraining.

Question 3

Does it profile members without anyone asking for it?

Accepted Answer

No, /profile is a moderator command that requires the Moderate Members permission. It scores 10 to 100 of a member's recent messages across scam risk, spamminess, noobness, toxicity and helpfulness, then assigns a persona and suggests next steps. Messages are cached in SQLite as they arrive, and if a member has no history you pick a channel to fetch recent messages on the fly.

Question 4

What do I have to enable in the Discord portal before any of this works?

Accepted Answer

Two privileged gateway intents, Message Content and Server Members, plus invite permissions covering Manage Messages, Moderate Members, Ban Members, View Channels, Send Messages, Embed Links and Read Message History. Application commands are deliberately not synced on startup, so you run the sync prefix command once as the owner, by DM or in a server channel.

Question 5

How does the punishment ladder actually escalate, and can I change the timings?

Accepted Answer

First offense is a warning DM, second a final warning DM, third a 10-minute timeout, and fourth and beyond 1-hour timeouts. /set-timeouts takes first_offense_mins and subsequent_offense_mins so an admin can change the third and later durations, /set-thresholds sets the tier1 and tier2 AI confidence sensitivity, and /user-offenses lists a member's infractions with resolution status.

Question 6

Which commands do moderators get, beyond /profile?

Accepted Answer

Moderators with Moderate Members get /user-offenses and an ephemeral /help reference; admins get /set-mod-log, /unset-mod-log, /set-timeouts, /set-thresholds, /pardon, /export-feedback in json or csv, and /mod-config, which reports the current settings and the active false-flag precedent count. You can also right-click a user, Apps, Generate AI Profile.

Question 7

What does it take to run this, and is there a test suite?

Accepted Answer

Python 3.10 or newer, a Discord bot token and a TypeSafe API key in .env, then clone, venv, pip install -r requirements.txt and python3 main.py, which creates the SQLite tables on startup. It is MIT-licensed, and python3 -m pytest -v covers migrations, message caching, channel scraping, the escalation ladder, admin view permissions and slash commands.

Jev Moderation Bot

About this project

Highlights

Quickstart

Questions about this entry

Reactions & coverage

From the Jev guides

From the Jev AI model community