Toxicity filter

The toxicity filter checks messages using the saved settings and can apply your chosen sanctions.

Configure and verify

Sign in, select the server and open Protection and select Edit beside Toxicity protection. April needs Kick Members and Ban Members for protection processing, along with permissions for the chosen action. Enable the filter, set its options, review global and filter role/channel permissions, choose Delete and/or Warn, then save. See Protection overview for setup and recovery.

The API projects a default of 90%; the Worker fallback is 70%, and values outside 70–101 become 90. A score at or above the effective threshold would match, with insult preferred over toxicity. However, the current dashboard form strips the Channels field while the API validator requires it, so a save can be rejected. Separately, the Worker compares that list to the member ID; an empty list or ordinary channel-ID list skips analysis before calling Perspective. This source trace has not been verified end to end, and channel-targeted toxicity filtering is not established as available. Provider errors or empty responses also produce no match.

If saving is rejected or no analysis occurs, do not rely on this filter for enforcement. Review the error and report the issue to the server administrator; use another appropriate moderation process while the limitation is investigated. Do not insert member IDs as a workaround.

The local edit form shows settings before saving; the filter was disabled.

Do not expect an ordinary channel test to establish a successful match until the setup limitation is resolved. Saved settings do not prove delivery or deletion. If the filter does not match, check its enabled state, permissions, exclusions and earlier matching filters. For a false positive, disable or adjust the rule and restore content manually if appropriate. Review automated actions before enabling escalation and messages before relying on warnings.