FabricLinx reads your network traffic at full fidelity — every packet, every flow, every microsecond. Detection happens at wire speed, not after a sampling window closes. From a single rack-unit appliance to carrier-grade 100G ingest.
AI agents that reason about networks have one structural problem: they assume the telemetry they consume reflects reality. Most monitoring stacks don't. They sample. They aggregate. They drop. By the time a packet loss event reaches the reasoning layer, the original packets are gone.
FabricLinx closes this gap. It sits between your network and the agents that reason on it, reading every packet, every BGP UPDATE, every syslog line, every optical level — at wire speed, with no sampling. By the time AutoLinx or SecureLinx forms a decision, the evidence is already preserved, indexed, and replayable.
FabricLinx is what makes "engineer-approved AI" practical. Engineers approve because the evidence is there. Auditors trust the ledger because the source data is intact.
For mid-market networks that need wire-rate insight without the cost of a carrier-class appliance. Single rack unit, fits in any branch or DC closet.
For enterprises running real production traffic and light SecOps workloads. Most common tier for banks, regional ISPs, and large mid-market customers.
For telcos, hyperscalers, and government carriers running aggregated 100G uplinks. Built to absorb the kind of traffic where sampling is not an option.
| Pairing | FabricLinx Lite | FabricLinx Pro | FabricLinx Carrier |
|---|---|---|---|
| AutoLinx software-only | — | — | — |
| AutoLinx + FabricLinx | ● | ● | — |
| SecureLinx (FabricLinx-native) | — | ● | ● |
| AutoLinx + SecureLinx, shared FabricLinx | — | ● | ● |
FabricLinx runs on programmable hardware — FPGA-based today (AMD/Xilinx Kria KR260 for Lite, Alveo class for Pro/Carrier). This is what enables wire-rate processing with deterministic latency.
The choice is intentional: software-only stacks on commodity NICs can't sustain 25 Gbps+ without sampling or dropping. We engineered FabricLinx so this isn't your team's problem.
For procurement: FabricLinx is a sealed appliance. You don't write Verilog; you consume an API.
Above FabricLinx, everything is open and software-defined: the agents, the LLM gateway, the audit ledger, the identity model, the vendor adapters.
FabricLinx outputs structured events — over gRPC, Kafka, or OTLP — consumable by AutoLinx, SecureLinx, or your own SIEM/SOAR if you have one.
You're not locked in. FabricLinx speaks open protocols. Your data stays yours.
When AutoLinx triages a packet loss event, FabricLinx provides the actual packets that were dropped — not a 30-second aggregate. The agent's root-cause analysis reaches optic-level findings (e.g. "eth1/3 optic at −8.2 dBm, below threshold") because the underlying data is intact.
When AutoLinx pushes a change, FabricLinx observes the resulting traffic shift within microseconds. If the change creates an anomaly, AutoLinx's rollback engine triggers before users notice. Without FabricLinx, validation depends on polling cycles.
SecureLinx detects East-West movement by analyzing flow patterns FabricLinx captures. When a detection fires, the evidence isn't a metadata snippet — it's the actual packets, replayable in the forensic console. Auditors and analysts can verify the call themselves.
SecureLinx compares firewall policy across PaloAlto, Fortinet, Cisco, and F5 — but more importantly, it compares the policy against the traffic FabricLinx is actually seeing. A rule that exists but never matches is a different problem than a rule that should exist but doesn't.
Every agent decision in the shared HyperLinxWork ledger links back to FabricLinx's packet store. Six months later, a compliance auditor can ask "show me what the agent saw when it made this call" — and the system answers with the packets, the flows, and the reasoning chain.
Today FabricLinx ships paired with AutoLinx or SecureLinx. On the roadmap: an API tier that lets teams using their own SIEM, SOAR, or in-house analytics consume FabricLinx events directly — without our agents.
If you're building a stack where FabricLinx stands alone, we'd like to talk.
Join the standalone design list →FabricLinx is being extended to support tenant-isolated traffic intelligence — for MSPs, telco wholesale, and datacenter operators who manage many customer environments from one platform.
In development with select design partners. Not GA. Capabilities and architecture co-designed with partners running real multi-tenant traffic.
Typical fit: MSP managing 10+ customer networks, telco wholesale provider, datacenter operator with multi-tenant service offering.
Apply as multi-tenant design partner →Add FabricLinx to an AutoLinx deployment for wire-rate diagnosis and instant change validation. Most teams choose FabricLinx Lite or Pro.
→ See AutoLinxSecureLinx ships FabricLinx-native. Pro or Carrier tier depending on aggregate throughput. Packet-level forensics included.
→ See SecureLinx