AUTOLINX · IN PRODUCTION Software-first NetOps · running with 3 operators across APAC · FabricLinx acceleration optional Start a pilot →
AUTOLINX · AGENTIC NETOPS · IN PRODUCTION SINCE 2025

Agentic AI for network
operations.

Engineer-approved by default. Reversible by design.

Describe a change once in plain English. AutoLinx translates your intent to Cisco, Juniper, Arista, Huawei, Nokia and 5 more vendor syntaxes — then executes in parallel across thousands of devices, only after engineer approval. Reversible by default. Every action evidence-backed and audited. Deploys software-first — add FabricLinx when you need wire-rate insight.

Our own AI · CPU-deployable 10+ vendors supported Engineer approval default Self-hosted · air-gap ready
INTENT → VENDOR SYNTAX · PARALLEL EXECUTION α-1 · watching
ENGINEER · WRITES INTENT ONCE $ intent upgrade BGP timers to 3/9 on all edge routers α AGENT α-1 · TRANSLATING Reading topology · 1,400 devices · 5 vendor families edge role identified · 247 routers match scope ↵ approve · 247 devices · rollback ready ✓ APPROVED Cisco IOS 62 devices router bgp 65001 timers bgp 3 9 Juniper 48 devices set protocols bgp hold-time 9 Arista EOS 54 devices router bgp 65001 timers bgp 3 9 Huawei VRP 38 devices bgp 65001 timer keepalive 3 Nokia SR 45 devices configure router bgp keepalive 3 +5 MORE VENDORS ✓ APPLIED · 247 routers · 5 vendor families · 1 intent · signed in audit ledger · rollback ready
1 intent · 5 vendor syntaxes · parallel execution · 100% audited ~ 10 seconds · intent → approve → applied
AUTOLINX IS THE AGENTIC EVOLUTION OF AUTOPROVISION
3 phases at a tier-1 Thai telco · 2019 → 2026
production-proven across 16-year relationship
Reference calls available under NDA during the briefing process. Request a reference call →
WHY AUTOLINX

NetOps that thinks. Engineers that decide.

Three reasons NOC teams ship AutoLinx into production — and keep it there.

Diagnosis under 60 seconds

root cause · not just symptom

AutoLinx reaches root cause — optic level, BGP state, CRC trend — not just the surface alert. Engineers get a proposal with diff, evidence, blast radius, and rollback path before they're fully awake.

Engineer-approved by default

agentic, not autopilot

Agents propose; engineers approve. Auto-approval is opt-in, scoped per action class, and always logged. Reversibility is non-negotiable — every change has a rollback that's been tested before push.

One intent grammar · 10+ vendors

Cisco · Juniper · Arista · Nokia · Huawei …

Describe the change once; AutoLinx translates to vendor syntax. No more porting playbooks across vendors. Compliance and audit work the same regardless of what's in your rack.

USE CASES

What AutoLinx does in production.

Three classes of work where customers measure clear MTTR and audit gains.
01 · TRIAGE

Wake the right engineer, with the answer attached

Median MTTR ↓ 85% · benchmarked at tier-1 Thai telco · validated across 3 anchor accounts

When an alert fires, the triage agent reads the relevant signals, forms a hypothesis, and proposes a fix — all before paging anyone. Engineers wake to a proposal with evidence, not an alert to investigate. Most nights, they approve and go back to sleep.

02 · CHANGE

Change management with rollback in the loop

Change failure rate ↓ 64% · 100% of changes carry a tested rollback

Engineers describe the change in intent. AutoLinx generates vendor-specific config, simulates blast radius, builds a rollback, and stages it for approval. After push, the agent watches for drift from expected state and offers to revert.

03 · COMPLIANCE

Continuous audit, ready to export

SOC 2 / ISO 27001 / PCI evidence packs auto-generated · 100% of actions captured

Every diagnosis, every change, every approval lands in a signed, append-only audit ledger. Compliance teams pull evidence packs directly — no scrambling before audit week. The ledger is vendor-neutral and replayable: six months later, you can replay what the agent saw and proposed.

ARCHITECTURE

Software-first. FabricLinx-optional.

AutoLinx ships as software. For most teams, that's complete. For higher throughput or wire-rate insight, FabricLinx drops in beneath without changing anything above it.
AUTOLINX AGENTS Triage · Change · Compliance reasoning · approval gates · rollback SHARED PLATFORM audit ledger · LLM gateway · identity · vendor adapters · evidence store INGEST PATH Software-only (default) SNMP · NETCONF · gNMI · syslog · BGP commodity infrastructure · cloud or VPC good for most teams · up to ~25 Gbps aggregate ▲ default for new deployments + FabricLinx (optional) wire-rate ingest · zero sampling · < 100µs FabricLinx Lite (10G) or Pro (40G) appliance when you need packet-level diagnosis ▲ add when you outgrow software-only
When to add FabricLinx
If your aggregate traffic exceeds 25 Gbps, or you need packet-level diagnosis (CRC chains, optic-level signal trend, sub-second loss events), drop in FabricLinx beneath AutoLinx. Nothing changes about the agent stack above — same APIs, same approval gates, same audit ledger.
Read about FabricLinx →
VENDOR COVERAGE

10+ vendors. One intent grammar.

Describe the change once. AutoLinx translates to vendor syntax — and supports the diff, blast radius, and rollback across all of them.
Cisco IOS-XE
✓ GA
Cisco NX-OS
✓ GA
Juniper Junos
✓ GA
Arista EOS
✓ GA
Nokia SR Linux
✓ GA
Huawei VRP
✓ GA
MikroTik
✓ GA
SONiC
✓ GA
Aruba CX
✓ GA
FRR / VyOS
✓ GA
START HERE

Four weeks. 100 devices. Measurable result.

Most teams pilot AutoLinx on a 100-device slice of their network — typically an edge zone or a regional NOC — with a solutions architect for setup. By week four, you have measurable MTTR and approval-throughput numbers from your actual environment.

No commitment to production. No pricing surprises. We either prove the gain in your environment, or we don't.

Pilot includes

✓ Up to 100 devices, your choice
✓ Solutions architect for setup
✓ MTTR baseline + measured outcome
✓ Read-only mode first 2 weeks
✓ Approval-gated changes weeks 3-4
Request a pilot →
QUESTIONS

What people ask first.

Does AutoLinx replace our existing NMS / observability stack?
No — and we don't recommend it. AutoLinx reads telemetry from devices directly and from your existing systems via adapters (PRTG, LibreNMS, SolarWinds, Datadog, custom). It adds reasoning and approval gates on top. Your existing dashboards keep working.
What if the agent makes a wrong call?
Every action is engineer-approved by default — wrong calls don't reach the network unless a human approves them. If they do, every change has a tested rollback path that AutoLinx is ready to invoke. False positives feed back into the agent baseline.
Can we self-host AutoLinx in our own VPC?
Yes. Self-hosted in your VPC is the default deployment. No configuration data leaves your network. AutoLinx runs our own AI stack — a purpose-built nano-LM + GNN reasoning model deployable on commodity CPU. For customers who want to augment with frontier models, we integrate with Bedrock, Vertex, Azure OpenAI, vLLM, or Ollama. Full air-gap deployments are supported on CPU alone.
When do we need to add FabricLinx?
Most teams don't, at least initially. Add FabricLinx when (a) your aggregate traffic exceeds ~25 Gbps and software ingest starts dropping samples, or (b) you need diagnosis at packet level — CRC trends, sub-second loss events, optic-level signal degradation. The agent layer stays identical above. See FabricLinx tiers →
How does AutoLinx differ from network automation tools like Ansible or Nornir?
Ansible/Nornir execute playbooks you write. AutoLinx reasons about what to do, then proposes the change for your approval. We integrate with both — AutoLinx can call existing playbooks as actions, and write its diffs back into your Git repos so the audit trail is consistent.
How is AutoLinx priced?
By device count, not by event volume or seat. Three tiers — Pilot, Production, Carrier-grade — with annual commit pricing. FabricLinx is priced separately when added. See full pricing →
GO DEEPER

Read AutoLinx at the level you need.

This overview sets the picture. Each of these pages goes one level further — into the technical architecture, the specific use cases, or the pricing structure.