Home/Platform
PathIQ™ AI Engine

Control Planes.
Not Control Rooms.
The Broadwan Platform.

A programmable 4-layer network stack that turns complex WAN operations into simple API calls. The network manages itself. You manage the business.

Stack Architecture

One Stack to
Replace Them All

Most enterprise WANs are 6–9 separate vendor products bolted together with duct tape and change tickets. Broadwan is a vertically integrated stack — from custom silicon at the bottom to AI at the top — all built and owned by one team.

Custom ASICBW-X1 (12.8 Tbps)
Forwarding planeP4-programmable
Control plane HA99.999% (5-nines)
Policy convergence<200ms global
AI training data500B flows / 8 years
Telemetry streams200+ signals/sec/link
🧠
L4 — PathIQ™ AI Inference
Congestion prediction · anomaly detection · closed-loop remediation
90s lookahead
🎛️
L3 — Intent Orchestration
SDN controllers · intent-to-config · BGP policy · RSVP-TE
<200ms
L2 — Programmable Data Plane
P4 forwarding · AES-256-GCM in hardware · hardware DPI · FEC
12.8 Tbps/chip
🌐
L1 — Physical Fabric
500+ PoPs · 6 undersea cables · 120 IXPs · owned fiber
40 Tbps
PathIQ™ AI Engine

Predict Congestion
Before It Exists.
Reroute Before It Hits.

PathIQ is trained on 8 years and 500 billion flows of network telemetry. It watches 200+ signals per second per link — RTT variance, loss patterns, BGP prefix churn, interface error counters — and predicts degradation 90 seconds before it materializes. Traffic moves while your path is still green.

Anomaly Detection
98.7%
True positive rate
Auto-remediation
91%
No human required
Prediction Window
90 sec
Before congestion occurs
Training Set
500B
Flows, 8 years
pathiq-monitor --live
# PathIQ continuous telemetry stream
[09:14:02] lon-fra-core-01: RTT delta +3.2ms · loss 0.01%
[09:14:08] PathIQ: BGP churn detected (AS7018) · confidence 84%
[09:14:09] PathIQ: Predicting congestion lon→fra in ~78s
[09:14:10] PathIQ: Pre-emptive shift: 40% flows → lon-ams-fra
[09:14:11] Tunnel health lon-fra: RTT 2.1ms (no impact)
[09:14:47] AS7018 route flap confirmed · lon-fra degraded
[09:14:47] User impact: 0 packets dropped · 0 sessions reset
» Incident closed autonomously. MTTR: 0 min.
»
Developer Platform

Declare Your Network.
Push. It Exists.

Every Broadwan resource is a first-class API object. Provision, configure, and observe your entire WAN with the same tools you use to deploy applications.

REST API v3

Full CRUD on Every Resource

Sites, circuits, tunnels, BGP policies, QoS profiles, analytics streams. 99.99% API SLA. SDKs in Python, Go, Node, Java, Rust.

GET /v3/sites?region=apac
POST /v3/tunnels
PUT /v3/policies/{id}
GET /v3/analytics/flows
DELETE /v3/circuits/{id}
Terraform Provider

Your Network as Code

Declare your entire WAN topology in HCL. GitOps-compatible. State stored in Terraform Cloud or your own backend.

resource "bw_site" "apac_hq" {
region  = "ap-southeast-1"
bw     = "10G"
qos    = "critical"
}
Kubernetes Operator

Network Policy in Your YAML

CRDs for every network resource. Deploy network policy alongside your application manifests. Works with Helm, Argo CD, and FluxCD.

kind: BroadwanPolicy
spec:
encryption: aes-256-gcm
qos: critical
ztna: enabled
FAQ

Technical Deep-Dive

How does BFD-triggered failover work under 50ms? +
Bidirectional Forwarding Detection (BFD) runs at 10ms hello intervals on all tunnels. When a link failure is detected, the BFD session drops immediately, triggering an instant route withdrawal in the local FIB without waiting for BGP hold-down timers. The BW-X1 ASIC executes the FIB update in hardware — the total failover path from detection to reroute is typically 30–48ms. ECMP pre-computation means the backup path is already installed before the failure occurs.
What happens to in-flight packets during a failover? +
For TCP sessions, most survive failover entirely via FEC (Forward Error Correction) which provides packet-level redundancy across paths. UDP flows (VoIP, video) use packet replication to secondary paths 10ms before the primary is declared dead. TCP sessions where packets are lost recover via normal TCP retransmission — typically under 200ms, invisible to most applications.
How does PathIQ know which path to route traffic to? +
PathIQ maintains a continuous multi-dimensional model of every link in the network — measuring RTT, jitter, packet loss, BGP route stability, interface error counters, CPU/memory on edge nodes, and 195 other signals every second. A gradient-boosted model predicts degradation probability for each path over the next 30s, 60s, and 90s windows. Traffic is steered to the path with the lowest predicted degradation probability for each application class, not just current observed quality.
Can I bring my own hardware for CPE? +
Yes. Broadwan runs as a vCPE (virtual CPE) on standard x86 hardware, most major hypervisors (VMware, KVM, Hyper-V), and cloud instances (AWS EC2, Azure VM, GCP GCE). Minimum spec: 4 vCPUs, 8GB RAM, 2x10G NICs. We also support physical CPE deployment on our purpose-built BW-100, BW-500, and BW-2000 appliances if you prefer vendor-managed hardware.
How does zero-touch provisioning actually work? +
When a new CPE is powered on, it makes a DHCP request and then calls home to the Broadwan cloud controller via an HTTPS bootstrap URL. The controller authenticates the device by its pre-provisioned serial number and X.509 certificate, then pushes the full configuration — routing policy, tunnel parameters, QoS maps, security policies — over an encrypted control channel. The device converges in under 15 minutes. You ship the box; your team plugs it in and walks away.
What observability data can I access from the API? +
Everything. NetFlow v9 and IPFIX records for all flows, per-application latency histograms, per-tunnel RTT/jitter/loss time-series at 10s resolution (1s on request), BGP routing table history, security event logs, CPE hardware health, and PathIQ confidence scores for every active path. All data is available via REST API, gRPC streaming, Kafka topic export, or direct S3 sink. Retention is 13 months at full resolution in our platform, or unlimited in your own storage.
Live Demo

See PathIQ Route in Real Time

Book a live technical demo — we'll show you PathIQ predicting and rerouting around a live outage in your region.

Book Technical Demo →