Problem
Most “managed hosting” is either white-labeled cPanel/Plesk or a custom UI over OpenStack/Kubernetes where one bad node breaks half the tenants.
Neither solves daily operations: provisioning, network debugging, backup recovery, process debugging, upstream escalation. A human still does this over SSH at 3 AM.
CoreSynth.io is the answer: hosting where an AI agent operates instances for you — sandboxed, auditable, under human supervision.
Stack
┌──────────────────────────────────────────────────────────────┐
│ WEB → Next.js 15 (App Router) + Tailwind 4 │
│ API → Node 22 + tRPC + Zod │
│ BILLING→ PHP 8.2 + custom (PDF invoices, CZ tax) │
│ DATA → PostgreSQL 15 + Redis (sessions, rate-limit) │
│ AGENT → Claude / OpenAI + MCP tool registry │
│ INFRA → Proxmox VE + Ansible + LXC + KVM │
│ EDGE → Nginx + Fail2ban + Crowdsec │
└──────────────────────────────────────────────────────────────┘Architecture
Three isolated layers, each with its own failure domain.
Results
- <90 s provisioning from order to SSH-ready instance
- 0 data loss over 18 months of operation
- 3 core nodes in CZ + DE, live migration
- 24/7 agent — resolves 90 % of alerts independently
- Currently in closed beta
What’s next
- CoreSynth v2 — new MCP-native agent, multi-region, self-service onboarding
- Public API — white-label for partners