Detect in <1s Nova
Nova screens every message through a RocketRide triage pipeline — sub-millisecond fast path — and quarantines threats instantly.
An AI-native company · Trust & Safety
The AI Trust & Safety team for every online community.
02 The problem
online communities
Discord, Slack, Telegram, forumstrust & safety market
spend on keeping people safe onlinehuman moderator cost per incident
and it doesn't scale with growthwhen scams, raids and crises hit
right when nobody is watchingVolunteer moderators burn out. Harm happens in seconds — response takes hours.
harm ~3s → response ~3h+03 The solution
Haven plugs into your community and runs a full Trust & Safety team of AI employees — detect, investigate, care, and learn — with a human founder approving every ban.
Nova screens every message through a RocketRide triage pipeline — sub-millisecond fast path — and quarantines threats instantly.
Rex detonates links in an isolated Celesto computer via an AdaL browser workflow: screenshots, redirects, forms, lookalike domains.
Mira — real face & voice via Masky.AI — answers crisis messages in seconds with 988 / findahelpline resources and pages a human via n8n. Crisis messages are never auto-deleted.
04 The team
Founder & CEO · approves every ban
Routes everything over BAND and writes the founder briefing.
T&S Triage Analyst
Threat Intel Investigator
Community Wellbeing Lead
Security Engineer
Data Analyst
Policy & Product Manager
Growth & Customer Success
05 How it works
Models via Flatkey fastreasoningcodingvision with failover and per-call cost tracking
06 Live demo moments
fake “free Premium” link in #general
flags risk 80+, hides it instantly
hands off to Rex over BAND
sandbox screenshot: credential harvester
voice warning to the channel
domain added to shared blocklist
n8n logs the enforcement
memberI can’t do this anymore. I want to end it.
MiraI’m really glad you told me. You’re not alone. Call or text 988, or visit findahelpline.com. A person from our team is joining now.
“Which kind of incident is most common?”
SELECT kind, count(*) FROM incidents GROUP BY 1 ORDER BY 2 DESC;
07 Safety & trust
Every ban needs founder approval — one click, with the evidence attached.
Crisis messages are never auto-deleted, always escalate to a human, and include 988 / findahelpline.
Names, emails, phones and IDs are redacted before any model call.
Community rules live as specs and every change is checked by Prelint.
No build ships with known-vulnerable dependencies: Snyk scans gate every merge, run on isolated Tenki CI runners.
ADMINISTRATORcriticaleval() on user inputcritical08 Sponsor stack
Every tool does a real job inside Haven’s workforce.
Workforce manifest for the 8 AI employees.
Real faces & voices for Mira, Rex and Aria.
Agent-to-agent messaging bus.
Onboarding, human paging, enforcement logs, weekly reports.
Triage pipeline with a <1ms fast path.
Browser workflow that drives link detonation.
Isolated sandbox computer for hostile links.
Tiered model gateway with failover + cost tracking.
Natural-language questions over live safety data.
Context compression for long incident threads.
Reusable SKILL.md skills learned from incidents.
Sandbox test runs & CI runners.
Spec & policy checks on every change.
Dependency vulnerability scans.
09 Business model & traction
Up to 1k members · Nova triage, Rex link scans, weekly report
Up to 25k members · the full 8-agent team, Mira crisis care, Ask Haven
Multi-community & platforms · custom policies, SSO, SLA
Unit economicsper incident
Market
MRRIllustrative projection
Traction
protected end-to-end by the Haven team today
10 Vision & ask
Discord, Slack and Telegram communities ready to run Haven alongside their mods.
Pilot with platforms and bot directories to protect communities at scale.
Credits and support to grow the AI workforce and onboard more communities.