Enterprise-grade performance for offices, pro-gamers and high-demand environments. A large-scale router built for those who refuse to compromise. Covers up to 7,500 sq ft in open environments and handles 400+ simultaneous devices — powered by the Qualcomm X75 5G module with GLOBAL SIM support.
The Rocket Pro uses the Quectel RM520N-GL — an industrial-grade 5G Sub-6 GHz M.2 module supporting both NSA and SA modes, with worldwide LTE-A coverage and speeds up to 3.3 Gbps downlink.
Every Rocket Pro that leaves us is loaded with our own custom build of OpenWrt SNAPSHOT (kernel 6.12, aarch64 Filogic 880). It isn't stock — we bake in a vendor-product modem dashboard so you can see the SIM, signal and Quectel RM520N-GL (Qualcomm X75) modem details at a glance, mwan3 dual-stack failover for IPv4 and IPv6, a 262,144-entry connection table, software flow offload, the CAKE queue manager for low-latency 5G, and the CPU pinned to the performance governor. It's the firmware equivalent of a service-and-MOT before the keys hit your hand.
Customers can request alternative firmware variants on a CD. These are signed Rocket Routers OpenWrt sysupgrade images, but flashing is at your own risk — an interrupted flash, the wrong image, or power loss mid-write can brick the device. If you'd rather not flash it yourself, get in touch and we'll do it before shipping or supply a pre-flashed second unit. Either way the 2-year warranty stays valid as long as the firmware came from us.
Every device, app and tab on your network opens "connections" — a Zoom call, a Spotify stream, every IoT sensor checking in. The router has to remember each one in a table called conntrack. A normal router ships with about 16,000 slots; once that table fills the router stops accepting new connections (it doesn't crash, it doesn't wipe itself — it just blocks new flows until existing ones time out, which is exactly the moment a busy office grinds to a halt). We've raised the ceiling to 262,144 entries with a 65,536-bucket hash so lookups stay fast under load — eight to ten times the headroom you'll find on a stock OpenWrt or consumer router. On top of that we enable software flow offload: once the kernel has decided how to forward an established connection, it drops the rest of the firewall walk and pushes packets through a fast in-kernel path, freeing the CPU for new connections, Wi-Fi, SQM and the rest. In the language of an enterprise network architect: long-lived TCP timeouts at 86,400 s, UDP at 30 s / 180 s, conntrack flushed on mwan3 link transitions to force re-pin, NAT/firewall offload (nft_flow_offload) running under nftables on Filogic 880's four Cortex-A53 cores pinned to the performance governor. In plain English: the router stops being the slowest thing on your network — even with 400+ devices on it.
Failover isn't a single ping check that flips a switch. mwan3 runs multi-target probes against 1.1.1.1, 8.8.8.8 and 9.9.9.9 (and the IPv6 equivalents 2606:4700:4700::1111, 2001:4860:4860::8888, 2620:fe::fe) on a 5-second interval. As soon as fibre fails three probes in a row it's marked offline, conntrack is flushed for that interface, and traffic instantly re-pins to wwan (your global SIM through the X75 modem). When fibre comes back up and passes its probes, traffic snaps back to the primary — again with a flush, so half-open TCP sessions don't sit awkwardly across two paths. Your IPv6 stack failovers in parallel through a separate fibre_then_lte6 policy so you don't lose v6 connectivity when v4 flips. End-to-end the failover is sub-10-second, and your team's TCP retries usually mask it entirely. For the people on the call, the Internet just stayed up.
The load-balancing variant uses the same mwan3 engine but a different policy: both wan_fibre and wwan are weighted active members. New connections are hashed across the two paths in proportion to the weight you set (default 80% fibre / 20% 5G — adjustable from LuCI). Existing connections stay sticky to whichever path they started on, so a long upload over fibre doesn't suddenly migrate mid-flight to a slower 5G session. If either path fails, the policy collapses to single-path mode automatically. Ideal for offices running cloud backups, persistent VPNs, video conferencing and SaaS at the same time, where you want to use every megabit you're paying for.
Click any image to enlarge
