[ Walkthrough ]
The agent reasons, the flexible context integration layer reaches your systems, and every answer is grounded in the data you actually run. It runs on your own hardware; nothing leaves the building.
An AI that understood your topology, configs and logs would turn hours of CLI work into minutes. The catch: that data can't leave the building, and rolling your own is a project most teams can't staff.
One agent, one data foundation: deep project work and fast operational answers, both entirely on-prem.
Segmentation reviews, port-security and hardening checks across 50 multivendor switches: network-wide, report-ready, in minutes.
"Which PLCs share a VLAN with the compromised HMI?" Blast radius and lateral-movement context for the SOC, without a CLI session per device.
Audit IT↔OT boundaries against IEC 62443 zones; validate against NIS2 and KRITIS hardening baselines.
"What changed at the substation in the last 24 hours?" Snapshot diffs and forensic timelines, ranked by severity.
Find every Siemens, Schneider and Moxa device by vendor and map it to its network location and zone.
Fast, targeted answers about devices, hosts and events: consolidated context instead of manual correlation across tools.
An agent that reasons, an open layer that connects, and a stack that runs entirely on your hardware.
The AI Stack is a standalone product, not an add-on. It talks to Explorer and Log Analytics today through the integration layer, and reaches third-party systems through the very same open layer.
The on-prem AI that works from day one, because the engineering most teams can't staff is done and packaged.
A GPU cluster designed to your requirements, the tuned open model, the context integration layer and the agent, built and calibrated together.
Through the flexible integration layer, you can connect your own systems, as well as Explorer and Log Analytics.
Assessments or incident questions: the agent calls the tools and grounds every answer in real data.
The model and the agent run on your hardware, inside your perimeter. No data goes to a cloud.
| Deployment | On-premises: a GPU cluster designed to your requirements; typically dual-node and scalable |
|---|---|
| Models | Open-source: GLM-4, Qwen 3, Devstral, Nemotron and others, tuned to the cluster |
| Integration | Flexible context integration layer |
| Live today | narrowin Explorer (digital twin), narrowin Log Analytics |
| Extensible to | Your CMDB and SIEM, monitoring, ticketing, SCADA/historian, firewall, via the integration layer |
| Interface | Chat, web client, API |
| Data | Nothing leaves your infrastructure: no cloud calls, offline-capable |
| Status | Beta: Explorer + Log Analytics integrations live |
No. The model and the agent run on your own GPU cluster, inside your perimeter. There are no API calls to a cloud provider; the system is offline-capable.
The integration layer between the agent and your systems. Each system is reached through a defined set of tools. Role-based access control (RBAC) keeps each user to what they are already allowed to access, so the agent never widens it. The layer is open: more systems can be added without changing the core.
Open-source models: GLM-4, Qwen 3, Devstral, Nemotron and others, selected and tuned to the on-prem hardware. Tool-calling quality matters more than raw size.
Knowledge questions can. Data-driven questions answered through the integration layer's tools stay grounded in your real configs, topology and logs, so every answer can be checked against the source.
That is the point of the open layer. Explorer and Log Analytics are live today; your CMDB, SIEM, monitoring and ticketing connect through the same integration layer.
No, it is a standalone product. It is a full, open on-prem AI in its own right, usually integrated with a customer's own systems. Standard integrations for Explorer, IPAM and Log Analytics are available out of the box.
The four-layer stack: assistant, local models, the context integration layer and the data it grounds on.
Read →
Using AI language models safely: what cloud LLMs put at risk, and what on-prem protects.
Read →
An Innosuisse project with ZHAW on proactive security and the vulnerabilities already inside the perimeter.
Read →Your network finally has something that can read all of it, reason over it, and never leave the building.
Talk to us about the AI Stack