Community Guard - Jev/Laya Spam Detection and Moderation

Community Guard uses System One models like Jev & Laya or traditional LLMs like Haiku to provide content moderation superpowers to community websites.

Features

• Screens the first few posts of new members, where nearly all spam comes from. Established members are never touched, and API costs stay tiny.
• Choose your engine: Jev through OpenRouter or TypeSafe, any OpenAI-compatible chat model (e.g. Claude Haiku), or any provider set up in the Drupal AI module. Laya coming soon (when it is available as a commercially hosted model - though I am open to adding local support)
• Confident spam is unpublished, never deleted. Repeat offenders are blocked, and everything they posted is taken down.
• One-click Unblock and restore allows moderators to restore anything a block action removed.
• Hides spammy profile fields, rate-limits bursts from brand-new accounts, and emails moderators on every block.
• A mobile-friendly verdict log with filters and a per-member view.
• A backtest command to compare engines and thresholds on your own history before switching.

Use it on forums, writing communities and membership sites where real accounts post promotional spam that CAPTCHAs and honeypots don't catch.

Post-Installation

1. Add your API key at Configuration » System » Keys.
2. Go to Configuration » People » Community Guard. Pick an engine and its key, describe your community in a few words, and choose which content types to screen.
3. Click Save and test the engine to see real verdicts, then tick Screen new members.

Screening runs on cron. Verdicts appear under Reports » Community Guard verdicts. Nothing is screened until you switch it on. Screened text is sent to your chosen provider, so mention that in your privacy policy.

Additional Requirements

• Key
• An API key for OpenRouter, TypeSafe or an OpenAI-compatible service. Or use the Drupal AI module (below) instead.

Recommended modules/libraries

• AI: use any configured chat provider (included submodule, experimental).
• TypeSafe AI Provider + AI Decision: run Jev through the AI module.

Similar projects

Honeypot and Antibot stop bots at the form. CleanTalk and SpamMaster score submissions against their own services.

Community Guard can be used by itself or in combination with other modules. It is tested in real world conditions to handles what gets past them.

Supporting this Module

Developed and used in production by Neopoet, to block what SpamMaster doesn't catch.