Real-Time Point-to-Multipoint Networking for Autonomous Systems and the Edge.

Where It Matters.

Speed, bandwidth, and stability where your connection is weakest.

SEND ONCE, POINT-TO-MULTIPOINT MESH DISTRIBUTION 1× EDGE UPLOAD · LIVE FAN-OUT & CONTROL RETURN

Send Once. Point to Multipoint.

Upload your live stream once to the nearest node. Orbtier fans it out only to active subscribers—whether one operator or twenty are watching, edge uplink usage stays flat.

Built for Real-Time. Prioritized on the Wire.

Nodes forward streams entirely in memory in under a millisecond, tagging critical control loops and live video with hardware DSCP priority ahead of bulk traffic.

Resilient Over Unstable Networks.

Standard protocols stall on packet loss and flood weak links with retries. Orbtier keeps live streams moving smoothly across lossy wireless links, multi-second dropouts, and power cuts.

Seamless Roaming. Zero Drops.

Move freely between Wi-Fi, LTE/5G, satellite, and local radios. When IP addresses or gateways change, active streams switch paths automatically without dropping.

Connect Everything. Trust Nothing.

Every device, payload, and network connected—backed by cryptographic security and full ownership.

Multi-Architecture. Cloud to FPGA.

One ~400 KB C11 daemon and static library running in 10.4–14.6 MB of RAM with zero runtime dependencies across x86_64, ARM64, ARMv7 FPGA SoCs, and MIPS routers.

Any Protocol. Bring Your Own Encryption.

Carry raw UDP payloads—video, sensor telemetry, ROS 2, MAVLink, or C structs—outbound-only through CGNAT and firewalls, with optional end-to-end payload encryption on top.

Cryptographic Access. Multi-Sector Isolation.

Every device authenticates with a key bound to an isolated logical Sector—segmenting robots, sites, or tenants on a shared mesh with zero cross-traffic.

Your Cloud. Your Network.

Deploy on your own cloud, edge hardware, or closed offline networks. No SaaS controller, no external login dependency, and zero telemetry leaving your infrastructure.

Measured, Not Simulated.

Verified on real Linux nodes, FPGA SoC boards, hardware routers, and continuous multi-day European internet runs.

52–210 µs

Node Forwarding Latency

In-memory relay median (p99 < 0.85 ms across 16 concurrent streams)

80,000 pkt/s

Peak Node Throughput

Single x86_64 node · 4,000 pkt/s MIPS · 1,000 pkt/s FPGA SoC

99.990%

216-Hour Multi-Arch Soak

414,720,000 packets verified across 6 European & LAN nodes

10.4 MB

Fixed RAM Footprint

Single thread · 9 file descriptors · zero runtime dependencies

View Full Hardware Benchmark Table & Wire Specifications
Metric Headline Figure Measured Context
Node Forwarding Latency 52–210 µs median (p99 < 0.85 ms) 16 concurrent streams (16,000 pkt/s each way, 0 lost of 960,000) & 5.04M datagram ladder
Peak Node Throughput 80,000 pkt/s (x86_64) · 4,000 pkt/s (MIPS) · 1,000 pkt/s (FPGA SoC) 2.4M datagrams (1,136 B) in 30 s at p99 742 µs; 548 µs median on a 4-core MIPS router; 1.05 ms median on an ARMv7 FPGA SoC board (dual Cortex-A9 @ 667 MHz)
Full MPEG-TS Video Framing 1,316 B (7 × 188 B) whole (0 fragmentation) 180,000 MPEG-TS datagrams across 30 pairs on 4 architectures (p99 within 1 ms of median on all cloud pairs; up to 1,952 B max delivery)
Continuous Multi-Arch Soak 414,720,000 packets (99.990% delivered) Three consecutive 72-hour internet + LAN rounds (216 hours continuous; 99.9983% across Round 3 cloud pairs) on amd64, arm64, mips, and an armhf FPGA SoC board (25.92M / 25.92M with 0 lost on the MIPS-to-FPGA wire)
Node Memory Footprint 10.4–14.6 MB RAM (1 thread, 9 FDs) Fixed resident memory from 10.4 MB (ARMv7 FPGA SoC) and 10.5 MB (MIPS) to 14.6 MB (x86_64) after millions of frames
Impaired-Link Overhead 0.0% added loss (+0.4 ms overhead) Tested under 15% packet loss, 200 ms delay, 50 ms jitter, 25% reordering, and 3,000 ms round-trip satellite-class delay
Crash & Sanitizer Verification 594 / 594 power-cut cells · 238 / 238 ASan/UBSan Survives power loss across 594 state-write phases and a real 6-node datacenter outage; zero-warning ISO C11 build

DiffServ / DSCP Wire Priority

Per-channel 6-bit DSCP tagging (0–63) preserved on both the inner frame and outer encrypted UDP packet across every hop.

Path MTU Probing & Framing

Built-in Don't-Fragment probing carries full 1,316 B (7 × 188 B) MPEG-TS frames unfragmented, up to 1,952 B max payload.

Cryptography & Sector Isolation

X25519, ChaCha20-Poly1305, Ed25519, and BLAKE2 tunnels with 128-bit sector_id isolation and startup binary manifest verification.

Multi-Architecture Binary Targets

Single-thread ISO C11 daemon (orbtierd) and static library (liborbtier.a) linked against standard libc alone on amd64, arm64, armhf, and mips.

Where We Connect.

One Robot. Remote Screens. One Video Stream.

Publish video and telemetry once from the robot; Orbtier delivers it simultaneously to remote drivers, engineers, recorders, and AI servers while returning prioritized control commands over the same link.

Local First. Connected When the Link Returns.

Run a node on a vehicle router or SBC so local devices keep streaming at full speed when satellite or cellular drops, then resync with the wider mesh automatically when connectivity returns.

Stream From Any Uplink Without a VPN.

Connect cameras, sensors, and edge computers outbound-only across LTE/5G, satellite, and CGNAT firewalls—no port forwarding, no virtual IP conflicts, and no VPN concentrator.

One Binary Replaces Five Tools.

Replace separate VPNs, message brokers, RTSP/WebRTC relays, NAT-traversal scripts, and certificate authorities with one lightweight C daemon and static library under your full control.