How it works
From alert to action in six auditable steps
ANE combines live telemetry, topology intelligence, retrieval-augmented reasoning, and an independent critic into one operational platform — deployable on-premises or in AWS, and built to keep a human in the loop.
The reasoning loop
How the AI brain thinks
Six steps from alert to action — every one auditable.
Alert arrives
A Grafana webhook fires.
Build context
Live BGP/OSPF state, the topology graph, and the relevant playbook corpus.
Propose action
The model selects a tool, arguments, and a confidence score.
Critic review
A second LLM independently verifies the proposal against live state.
Operator approval
A human approves or rejects — the loop stays human-in-the-loop.
Execute & verify
Changes run with checkpointing, verification, and automatic rollback.
AI Chat
Natural language over live network state
Ask the network anything. Get grounded answers with citations.
- Live-state synthesis. “What is R1’s status?” returns real BGP/OSPF/interface state, not stale data.
- Blast-radius analysis. “What if XR1 goes down?” returns graph-traced impact.
- Corpus-cited answers. Every claim links to the playbook chunk that supports it.
- Explainable decisions. “Why did the critic reject this?” returns the model’s own reasoning.
- Demo presets. One-click, pre-canned queries for repeatable demonstrations.
- Identify the flapping peer
- Check for a hold-timer mismatch
- Verify the OSPF underlay is stable
- Check interface error counters
Deployment · on-premises
A single appliance for PoC; Kubernetes for carrier-grade scale
The same platform, sized for the job — from a one-day proof of concept to a high-availability cluster.
| Dimension | PoC / small operations | Production / carrier-grade |
|---|---|---|
| Footprint | Single Linux VM | HA Kubernetes cluster |
| Compute | 8 vCPU / 32 GB RAM | 3+ nodes · 16 vCPU / 64 GB each |
| Storage | 200 GB SSD | Ceph or NFS-backed PVs |
| Network | 1 NIC (management) | Out-of-band + in-band NICs |
| Identity / OS | Ubuntu 22.04 / RHEL 9 | Active Directory / SAML / OIDC |
| LLM | External API (Claude / OpenAI) | On-prem (Ollama / vLLM) or external |
Ideal for highly regulated environments, air-gapped networks, and strict internal security requirements.
Deployment · AWS
Managed services where they save toil; EC2 where they don’t
A multi-AZ reference architecture that keeps your data inside your own AWS account.
Backend, MCP, webhook, and RAG services.
TLS, WAF, and Cognito at the edge.
Claude Sonnet in-region. Data never leaves your AWS account.
Vector embeddings; Timestream for telemetry.
Dashboards and alerting.
ElastiCache (Redis event bus), RDS PostgreSQL (NetBox / audit), and S3 (corpus / checkpoints).
Secure link to your on-prem management network.
Ideal for rapid deployment, elastic scalability, and managed-services integration.
Next
See where it’s going.
Today ANE is at Phase 8 — and the roadmap is well-defined.
View the roadmap →
Let's Connect
Share Your Netwwork profile and we’ll Build an ROI Model and a Working Demo of ANE against your Vendor Mix.
By providing a telephone number and submitting the form, you are consenting to be contacted by SMS text message and agreeing to our Privacy Policy. Message frequency may vary. Message and data rates may apply. Reply STOP to opt out of further messaging. Reply HELP for more information.