What-If Network Failure Analysis with CogniNet
Model link, device, and interface failures before they happen. CogniNet's GNN predicts blast radius, convergence time, and affected services for any proposed change or failure scenario.
CogniNetBatfishNeo4j
PyATSThe problem today
Network teams discover blast radius during outages, not before them. Manual what-if analysis means opening a spreadsheet, tracing paths by hand, and hoping you remembered every ECMP link and FHRP failover group. When a change review asks 'what happens if this link goes down?', the honest answer is usually 'we think it's fine' — backed by tribal knowledge, not data. Post-mortems routinely reveal failure paths nobody modelled.
How AI agents solve it
CogniNet builds a heterogeneous graph of your entire network — devices, interfaces, subnets, VRFs, links, protocols, security zones — and runs Graph Neural Network inference across it. Feed it a failure scenario (link down, device offline, interface flap, cost change) and it returns outage probability, blast radius as a list of affected nodes and services, estimated convergence time, and the rerouting paths traffic will take. The 7 specialized analyzers (routing, L2, FHRP, multicast, security, capacity, convergence) validate that the failover paths are actually viable, not just theoretically present.
Who this is for: Network engineers, change advisory boards, and SRE teams responsible for network reliability
Manual workflow vs. CogniNet Agent
Manual workflow
- What-if analysis means tracing paths on a whiteboard
- Blast radius is discovered during outages, not before
- FHRP and ECMP failover paths are assumed, not validated
- Change reviews rely on tribal knowledge
- Post-mortems routinely find failure paths nobody modelled
With the CogniNet Agent
- Any failure scenario modelled in seconds with full graph analysis
- Blast radius quantified before changes go live
- Failover paths validated by 7 specialized analyzers
- Change reviews backed by GNN-computed evidence
- Pre-emptive remediation for weak failover paths
How the CogniNet Agent runs this
- 01
CogniNet Agent ingests current network configurations from all vendors (Cisco IOS/NX-OS/ASA, Palo Alto, F5)
- 02
Graph builder constructs a heterogeneous PyTorch-Geometric graph with 70+ edge types
- 03
Operator selects a failure scenario: link failure, device failure, interface failure, or cost change
- 04
GNN model runs forward pass with the failure injected into the graph
- 05
7 analyzers validate failover viability: OSPF reconvergence, BGP path selection, FHRP promotion, STP topology change
- 06
Results surface: outage probability, blast radius nodes, convergence time estimate, rerouted paths
- 07
Remediation engine generates pre-emptive fixes if the failover path has known weaknesses
Measurable impact
Eliminates surprise blast radius during outages
Change approval backed by computed failure analysis, not guesswork
FHRP and ECMP failover gaps caught before they matter
Convergence time estimates inform maintenance window planning
Agents involved
Part of our Network Visibility solution
This use case is one piece of a larger pipeline
CogniNet what-if analysis is part of the Network Visibility solution — see the full pipeline from discovery to cognitive analysis.
Explore the Network Visibility solutionGoverned by the AI Gateway
Every agent action in this use case is audited, policy-checked, and cost-tracked
Structura's AI Gateway sits between every agent and the underlying LLM providers. Every decision made during this use case. Every plan review, every policy check, every fix PR, is routed through guardrails, logged to an immutable audit trail, and evaluated against NIST AI RMF and AIUC-1 controls.
Learn about the AI GatewayRelated use cases
Keep automating
Network Cognition Scoring for Executive Reporting
A single weighted score — the Network Cognition Score (NCS) — that captures performance, confidence, risk, impact, and intent alignment across your entire network, with executive summaries your CTO can read.
Intent-Based Network Validation with CogniNet
Define your network intent in YAML — routing policy, redundancy requirements, security posture — and let CogniNet validate that your actual configurations match your design intent, with remediation for every deviation.
Multi-Vendor Network Configuration Analysis with CogniNet
Parse and analyze configurations from Cisco IOS, NX-OS, ASA, Palo Alto, and F5 through 7 specialized analyzers and a 108+ rule engine — in a single pass.
See this use case in a live demo
We'll walk you through exactly how the CogniNet Agent handles this in a real environment with your stack, your policies, and your constraints.