Skip to content

Trust — infrastructure

This is the list of servers that actually handle your traffic or your files. Every IP below is a public endpoint. You can ping, traceroute, or whois any of them and the result should match what’s on this page.

Katafract LLC owns the accounts on every provider. The providers host the hardware; we operate it. No third-party managed services sit in the data path for VPN, DNS, or file storage.

Each node runs WireGuard (wg1, UDP 51821) and Haven (AdGuard Home bound to the WG interface). Peer isolation is enforced server-side: clients can reach the internet but not each other and not the mesh.

NodeRegionProviderIPv4WG client CIDR
vpn-nbg-01Nuremberg, DEHetzner178.104.49.21110.11.1.0/24
vpn-hel-01Helsinki, FIHetzner204.168.224.24310.11.2.0/24
vpn-pdx-02Hillsboro, OR, USHetzner5.78.207.19910.11.3.0/24
vpn-iad-01Ashburn, VA, USHetzner87.99.128.15910.11.6.0/24
vpn-pdx-01Hillsboro, OR, USHetzner5.78.178.20210.11.7.0/24
vpn-sin-02Singapore, SGVultr149.28.132.18410.11.8.0/24
vpn-sin-03Singapore, SGVultr45.76.186.7510.11.9.0/24
vpn-nrt-01Tokyo, JPVultr167.179.82.21610.11.10.0/24
vpn-bom-01Mumbai, INVultr65.20.76.5610.11.12.0/24
vpn-ewr-01Piscataway, NJ, USVultr64.176.215.9610.11.13.0/24

GeoDNS routes the vpn-*.katafract.com names to these addresses. The mapping is published under GeoDNS records and each name resolves to exactly one IP above.

Vaultyx ciphertext chunks land here. Replication factor is 2 across zones.

NodeRegionProviderRoleCapacity
furySt. Louis, MO, USContaboGarage node, zone us-central1.5 TB
atlasVint Hill, VA, USOVH KimsufiGarage node, zone us-vin14.8 TB
hadesBeauharnois, QC, CAOVH KimsufiGarage node, zone ca-bhs14.8 TB

Public S3 traffic reaches the cluster through s3.objstore.io (proxied via argus nginx). Direct Garage ports are not exposed to the internet.

One box runs a warm standby for the control plane and the monitoring stack. It does not carry user VPN or DNS traffic.

NodeRegionProviderRoleIPv4
furySt. Louis, MO, USContaboartemis-api standby, Prometheus, Grafana85.239.240.208
  • Hetzner — EU + US nodes. Cloud + dedicated.
  • Vultr — APAC + US-East nodes. Cloud.
  • Contabo — US Central VPS (Missouri) used for monitoring + Shards us-central.
  • OVH Kimsufi — North American dedicated storage boxes (atlas, hades).
  • Self-hosted at Tek’s home — a mini PC (tartarus) runs a standalone Garage cluster used only by the founder as a dogfood customer. It is not part of the shared Shards cluster and holds no other user’s data.

The internal control plane (artemis, argus, kata-db-replica) and the mesh-only automation infrastructure don’t appear in this table. They never handle user VPN traffic and never hold plaintext user content. Their topology is described in platform architecture overview.

This table reflects the live fleet. It is updated each time a node is added or retired. Refresh is triggered from the katafract-audit pipeline that inspects the real node registry on artemis, so drift between this page and reality shows up in the next commit.

We add nodes continuously. Treat the list as “what’s live right now,” not a ceiling. When a new region comes online it will appear here before it appears in the client app’s server list.

Pick any IP above and run:

whois <ip> | grep -iE 'orgname|netname|country'
mtr -r -c 5 <ip>

The whois result should identify one of the providers listed. The mtr result should terminate at the IP, not a CDN edge.

  • Logs policy — what is and isn’t captured on these nodes
  • Retention — how long anything that is captured lives
  • Threat model — what this fleet is designed to defeat