What We Do

Rule-Based Detection

Every message passes through a multi-method content detection engine — no external AI dependency, no per-message API cost, no single point of failure.

• Keyword matching — case-sensitive or insensitive, whole-word or substring
• Regex pattern matching against configured rulesets
• Content fingerprinting of known harmful material
• URL pattern matching against domain and path blacklists
• Each rule carries a severity score; the score determines the automatic response

Detection completes in under 50ms from message received to action taken.

Adaptive Learning

Every community has different norms. Guardiobot builds a behavioural profile per server and flags content that meaningfully deviates from it — reducing false positives without sacrificing coverage.

• Tracks message length, emoji rate, link frequency, caps percentage, and mention patterns
• Sampling intensity is configurable: passive, active, or aggressive per server
• Low-severity detections feed directly into the learning system
• Profiles are updated on a regular schedule — not in real time — to prevent manipulation

Message samples are automatically deleted after 30 days.

Cross-Server Enforcement

A ban issued in one server propagates automatically across every server in the Guardiobot network — bad actors cannot escape by server-hopping.

• Every member join triggers a live check for active cross-network punishments
• Bans from report actions and automated detections are both propagated
• Network participation is opt-in per server, with configurable trust thresholds
• Low-confidence bans go to a manual review queue before enforcement

Network settings are managed exclusively by Guardiobot staff — server owners cannot modify network participation directly.

Transparent Operations

Clear documentation of how our systems work, what data we collect, and how long we keep it. No vague policies.

Read our Privacy Policy — full details on exactly what is stored and for how long.

GDPR & CCPA Compliant

Privacy compliance is built into the architecture, not added afterwards.

• Sensitive data is cryptographically hashed at write time — never stored in plaintext
• Message samples purged after 30 days, audit logs after 90 days
• Right to erasure and right to data portability both implemented
• Deleted users retain only the minimum necessary data for ongoing ban enforcement

Appeal System

All bans are appealable by default. Appeals are reviewed by senior volunteers and tracked in the audit log.

• Every moderation action records whether the action is appealable
• Users are notified of their appeal status via direct message
• Senior volunteers can restrict or restore appeal rights on a per-case basis

Submit and track appeals through your dashboard.

Human Review Process

Automation handles the clear-cut cases. Humans handle everything else. All volunteers are verified and credentialled.

• All volunteers are publicly verifiable at guardiobot.org/verify/{id} — no login required
• Volunteer credentials use industry-standard password hashing
• Credentials expire annually and must be actively renewed
• Entry-level volunteers cannot see usernames or reporter identities — redacted at the API layer, not just the UI
• NSFW reports are restricted to volunteers with verified age and signed legal documentation

High Reliability

Automated health monitoring, connection pooling, and background maintenance keep protection running without gaps.

• Health checks monitor all core services continuously
• Database connection pooling with automatic recovery on failure
• Background tasks handle housekeeping automatically — expired punishments, stale sessions, rate limit cleanup

View live operational status on our status page.