AI-Powered Scam Detection — Now with Custom Rules Thrella's AI detection system got a major upgrade. You can now create multiple AI detection rules, each with their own prompt, action, confidence threshold, and channel scope. Whether you want strict moderation in your trading channels but lighter checks in general chat — you can set that up per rule. You can also write custom AI prompts tailored to your server's context, and test them right from the dashboard before going live.

Smarter Escalation System Repeated offenders now get handled automatically. Set up escalation rules that stack — if a member keeps violating filters, Thrella can automatically escalate from a warning to a timeout, kick, or ban. Each step triggers independently so you stay in control of the severity ladder.

Dashboard Case Quick Actions Managing cases just got faster. You can now ban, kick, timeout, or warn members directly from the cases tab — no need to switch to Discord. The moderation page has been reorganized into clean tabs for Cases, Appeals, Mod Actions, and Settings.

Better Link Detection We've hardened Thrella's link filter against several obfuscation tricks. False positives on YouTube links and other legitimate URLs have also been fixed.

Per-Hash PFP Actions The profile picture filter now supports per-pfp action overrides. If a specific scam avatar keeps showing up, you can assign it a harsher action (like an instant ban) while keeping the default filter action lighter for other matches.

Public Filter Notifications Optionally let your members know when Thrella takes action. Per-filter public notifications post a brief message in the channel where a violation happened, with auto-delete after a configurable delay. Available for link, invite, forward, block-all-URLs, and AI detection filters.

Dashboard Access Control (RBAC) Grant trusted moderators view-only or edit access to specific sections of the dashboard — settings, security, cases, billing, AI detection, and more. View-only users see everything but can't change configs.

Audit Log Improvements

Audit logs show the actual actor who deleted a message (not just "message deleted")

New dedicated MEMBER_WARN and MEMBER_KICK events for clearer mod action tracking

Richer log details: sticker previews, attachment thumbnails, role change badges, and parsed custom emotes

Allowlist Pattern Testing The domain allowlist now supports glob patterns (wildcards like *.example.com), and includes a built-in URL test tool so you can verify your patterns match what you expect before saving.

Quality of Life

AI image scanning now resizes images before analysis — lower token costs, same accuracy

Filter deduplication prevents the same message from triggering multiple actions

Forum channels are now properly supported in AI detection channel exclusions

Numerous false-positive fixes across all detection engines