1. Evolution of the Marzban Family
Marzban popularized managing Xray-core via its native gRPC API inside a clean Python/FastAPI backend. As operators scaled to multiple countries, Marzneshin emerged to decouple the central dashboard from distributed Marznode workers (supporting both Xray and Hysteria 2). PasarGuard pushed further into commercial reseller workflows, splitting Control Plane (Python/React) and Data Plane (Go nodes) over Docker and Caddy.
2. Failure Modes: Central Control Plane vs Autonomous Edge
Centralized master-node panels optimize for a single global user table, but create operational fragility:
- Control Plane bottleneck: If the master server, PostgreSQL/MySQL database, or Caddy/Docker ingress goes offline or gets blocked, provisioning and token sync across all edge nodes stall.
- Heavy stack maintenance: Managing Docker Compose, Python dependencies, reverse proxies, and gRPC node certificates across regions raises debugging complexity.
- Zero clientless support: None of the three platforms manage native Windows SSTP or hardware-router L2TP/IPsec tunnels.
Marzban, Marzneshin & PasarGuard vs Apium Cloud OS
Architecture ComparisonComparing orchestration complexity, database dependency, and protocol reach.
| Criterion | Marzban | Marzneshin | PasarGuard | Apium Cloud OS |
|---|---|---|---|---|
| Control Stack | Python (FastAPI) | Python + Marznode | Python/React + Go + Caddy | Compiled Rust |
| Deployment | Docker / Python | Master + Edge Nodes | Docker Control/Data Plane | Zero-Disk SSH single binary |
| Node Autonomy | ◐ Single-host | ✕ Depends on Master | ✕ Depends on Master | ✓ 100% Autonomous |
| Cascade Failover | ✕ No | ◐ Node list only | ◐ Node list only | ✓ Upstream Relay + Failover |
| SSTP & L2TP/IPsec | ✕ No | ✕ No | ✕ No | ✓ Native built-in |
When Marzban / Marzneshin / PasarGuard Fit
- You operate a large reseller or Telegram subscription network needing one shared user database across 20+ proxy nodes.
- You already have DevOps workflows built around Docker, FastAPI, and Marzban billing bots.
When Apium Cloud OS Wins
- You want zero central DB single point of failure—every gateway node runs autonomously and survives even if management links drop.
- You need corporate L2TP/IPsec routers, Windows SSTP, OS-level nftables/tc control, and multi-hop relay failover.