# Warburton

> The Community Operating System for Online Communities.

Warburton is the Community Operating System for online communities, helping human moderators govern safer, healthier and more sustainable communities through AI-assisted governance, operational assurance and community memory. It runs inside Telegram, Discord, and WhatsApp, providing governance, safeguarding, knowledge capture, operational intelligence, community health monitoring, welfare escalation, deterministic safety, daily digest, and contributor recognition — deployed as a single platform with a consistent character voice. 4,183 passing tests. Every decision auditable.

## Pricing

- $727 per month plus tax — 24/7 community coverage for less than $1/hour
- 14-day free pilot included
- No setup fees, no long-term contract required
- Price protected until October 2027

## Real Deployments

Two verified case studies, details confirmed and cleared for publication by the people involved — not hypothetical scenarios. Full write-ups: https://iamwarburton.com/case-studies

- **Veterans community (welfare/contextual safety):** A private community for military veterans communicates largely through dark, self-deprecating humour as a normal register. Warburton tracked that baseline and recognised when a member's language shifted into a genuine crisis. Two welfare referrals were made; one was subsequently referred for further psychological therapy.
- **Founder's holiday (coverage/operational confidence, ~500-member Solana memecoin community):** A founder who hadn't taken a real break in two years took his first holiday after gaining confidence in Warburton's coverage. A scam attempt, an antisemitic comment, and an Islamophobic pile-on were all caught and handled while he was away; he extended the trip by another day once he saw the community was genuinely fine without him.

## Pages

### Overview
- What is Warburton?: https://iamwarburton.com/what-is-warburton
- How Warburton Works: https://iamwarburton.com/how-warburton-works
- Why Warburton: https://iamwarburton.com/why-warburton
- About: https://iamwarburton.com/about
- FAQ (17 questions): https://iamwarburton.com/faq
- AI Governance: https://iamwarburton.com/ai-governance
- Architecture: https://iamwarburton.com/architecture
- Case Studies (2 real, 3 illustrative): https://iamwarburton.com/case-studies
- Testimonials: https://iamwarburton.com/testimonials
- Press & Media Kit: https://iamwarburton.com/press
- AI Community Management: https://iamwarburton.com/ai-community-management

### Problem-led guides
- Stopping Scams and Pile-Ons: https://iamwarburton.com/stopping-scams-and-pileons
- Detecting Community Distress: https://iamwarburton.com/detecting-community-distress
- Preventing Moderator Burnout: https://iamwarburton.com/preventing-moderator-burnout

### Capabilities
- Community Health Monitoring: https://iamwarburton.com/community-health
- Operational Assurance: https://iamwarburton.com/operational-assurance
- Protection Protocols: https://iamwarburton.com/protection-protocols
- Knowledge Capture: https://iamwarburton.com/knowledge-capture
- Welfare Ladder: https://iamwarburton.com/welfare-ladder
- Governance: https://iamwarburton.com/governance
- Daily Digest: https://iamwarburton.com/daily-digest
- Deterministic Safety: https://iamwarburton.com/deterministic-safety
- Contributor Recognition: https://iamwarburton.com/contributor-recognition

### Comparisons
- Warburton vs Discord AutoMod: https://iamwarburton.com/warburton-vs-discord-automod
- Warburton vs MEE6: https://iamwarburton.com/warburton-vs-mee6
- Warburton vs Dyno: https://iamwarburton.com/warburton-vs-dyno
- Warburton vs AI Moderation Tools: https://iamwarburton.com/warburton-vs-ai-moderation
- Warburton vs Traditional Moderation: https://iamwarburton.com/warburton-vs-traditional-moderation

## Canonical Definitions

- **Community Health Monitoring (COS-001):** Real-time community temperature monitoring, silence detection, repeated-target tracking, moderator burnout detection, positive community metrics, and exclusion pattern identification.
- **Operational Assurance (COS-002):** Decision provenance, circuit-breaker alerting, and Prometheus metrics with Grafana dashboards — every routing chain persisted from trigger to outcome.
- **Protection Protocols (COS-003):** Three deterministic protection protocols with no AI judgement in the loop — pile-on detection, absent-person protection, and pattern-break detection.
- **Knowledge Capture (COS-004):** Provenance-weighted knowledge cache attributing answers to the humans who gave them, with a 71% cosine similarity quality standard and quarterly promotion gates.
- **Welfare Ladder (COS-005):** Per-user sentiment tracking over time with multi-rung escalation, crisis-level human intervention, and a full welfare audit trail.
- **Governance (COS-006):** Consent ledger with 2,562-day FCA-aligned retention, programmable policy engine, 9 kill switches, hash-chained audit trails, and board-reportable incident governance.
- **Daily Digest (COS-007):** Daily or weekly digest delivered to admin DMs — unanswered questions, moderation actions, sentiment shifts, new members, and quiet departures. No dashboard login required.
- **Deterministic Safety (COS-008):** Eight hard rules with no AI in the loop — persona lock, voice anchor, citation guard, repetition guard, fake-fetch intercept, fresh time injection, and related safety modules.
- **Contributor Recognition (COS-009):** Federated threat intelligence with pseudonymous contributions across deployments, contribution confidence scoring, and automatic promotion recommendations.

## Architecture Overview

Warburton is structured in three workstreams:

- **WS1 — Platform Core:** Telegram, Discord, and WhatsApp adapters, character engine, message pipeline, and tiered multi-model response generation (local Ollama → Granite fallback → Claude escalation) with cross-model quality judging. SSRF-protected fetch pipeline with per-hop private-IP validation and DNS rebinding protection.
- **WS2 — Community Operating System:** The nine capability areas (COS-001 through COS-009). Backed by a comprehensive AI attack surface audit covering six categories (tool misuse, permission boundaries, memory poisoning, workflow manipulation, cross-session persistence, data exfiltration) with residual risks documented and severity-rated.
- **WS3 — Operations & Governance:** Full decision provenance via Context Accumulator Pattern (12 fields per response, 12 decision-reason categories across 21 return points). 19 recordSafetyTrigger call sites across 5 moderation modules. Prometheus metrics, Grafana dashboards, circuit-breaker alerting with health checks for Ollama/Claude/Redis. Consent ledger, hash-chained audit trails, and the governance layer. Tenant isolation with SameSite=Strict CSRF cookies (constant-time comparison) and atomic OAuth state consumption. Production/staging/test environment segregation aligned with ISO 27001 practice. Action-truth principle: a moderation action is never reported as taken unless the underlying platform call actually confirms success.

## Capability Areas

### Community Health Monitoring (COS-001)
Real-time community temperature monitoring, silence detection, repeated-target tracking, moderator burnout detection, positive community metrics, and exclusion pattern identification.

### Operational Assurance (COS-002)
OA1 decision provenance via Context Accumulator Pattern — 12 fields per response including routing_path, model_used, retrieval_sources, middleware_intercepts, token counts, cost, and decision_reason (12 deterministic categories across 21 return points). OA2 circuit-breaker alerting with admin alerts for welfare/crisis events, budget exhaustion, and provenance write failures; health checks for Ollama, Claude, and Redis. OA3 safety pipeline metrics — 19 recordSafetyTrigger call sites across 5 moderation modules (scam, NSFW/advances, safeguarding, community standards, welfare/distress) with WhatsApp adapter parity.

### Protection Protocols (COS-003)
Three deterministic protocols with no AI judgement in the loop. PP1 pile-on detection (10-stage escalation ladder), PP2 absent-person protection, PP3 pattern-break detection (deterministic state machine). Enforced identically across Telegram, Discord, and WhatsApp via fixed-order safety pipeline including edit handlers. Every detection metered to Prometheus via recordSafetyTrigger and recordModeration.

### Knowledge Capture (COS-004)
Five knowledge domains (learned community answers, Substack diary, specialist lore, Wikipedia, Reddit) with three retrieval tiers (cache → local Ollama → Claude escalation). Provenance-weighted cache attributing answers to the humans who gave them. 71% cosine similarity quality standard. Quarterly promotion gates. 90-day TTL on learned cache with 500-entry FIFO cap. Opus answer seeding for unanswered questions, clearly marked as AI-sourced. Write-time and read-time injection guards on all retrieval paths (defence-in-depth) using 8 canonical detection patterns.

### Welfare Ladder (COS-005)
Per-user sentiment tracking over time. Multi-rung escalation with defined criteria at each level. Crisis-level intervention always routes to a human operator. Full welfare audit trail.

### Governance (COS-006)
Consent ledger with 2,562-day FCA-aligned retention. Programmable policy engine. 9 kill switches. Hash-chained audit trails (audit logging never throws — retries without caller's transaction on failure). Incident governance with PostMortem, Timeline, and Operations views. Board-reportable by design. AI attack surface audit covering 6 categories (tool misuse, permission boundaries, memory poisoning, workflow manipulation, cross-session persistence, data exfiltration) with residual risks documented and severity-rated. Tenant isolation enforced per-deployment across all data paths.

### Daily Digest (COS-007)
Daily or weekly digest delivered to admin DMs. Surfaces unanswered questions, moderation actions, sentiment shifts, new members, and quiet departures. No dashboard login required.

### Deterministic Safety (COS-008)
Eight hard rules with no AI in the loop: persona lock, voice anchor, citation guard, repetition guard, fake-fetch intercept, fresh time injection, injection guard (8 canonical detection patterns applied at every LLM context entry point), and memory poisoning guard (write-time sanitisation + read-time filtering with 200-char extraction cap). Fixed-order safety pipeline enforced across all three platform adapters including edit handlers. Cross-model quality judging validates safety constraints across different model behaviours.

### Contributor Recognition (COS-009)
Federated threat intelligence with pseudonymous contributions across deployments. Contribution confidence scoring. Automatic promotion recommendations. What one community learns, others benefit from.

## Platforms

- Telegram
- Discord
- WhatsApp

## Built By

Costwolds LLC
Website: https://costwoldsllc.xyz

## Links

- Website: https://iamwarburton.com
- Company: https://costwoldsllc.xyz
- Newsletter: https://thewarburtondiaries.substack.com
- X / Twitter: https://x.com/warburtondiary
- Instagram: https://instagram.com/bernardsundayshoot
- Press & Media Kit: https://iamwarburton.com/press

## Press Coverage

- "Warburton Launches Governance and Incident Continuity Infrastructure for Online Communities" — openpr.com, May 25, 2026: https://www.openpr.com/news/4523674/warburton-launches-governance-and-incident-continuity
- "AI Community Platform Warburton Begins Appearing in Google AI Overviews" — openpr.com, 2026: https://www.openpr.com/news/4593878/ai-community-platform-warburton-begins-appearing-in-google-ai
- "Warburton AI Community Manager Demonstrates Safeguarding Potential in Veteran Community Pilot" — prlog.org, August 17, 2026: https://www.prlog.org/13165152-warburton-ai-community-manager-demonstrates-safeguarding-potential-in-veteran-community-pilot.html

## Contact

Visit https://costwoldsllc.xyz or subscribe to https://thewarburtondiaries.substack.com for updates.