The Architecture: AI Beside the Database
The pattern is straightforward: a service deployed on a VM inside your network connects to the SyteLine SQL Server over internal networking, builds its knowledge layer there (schema index, procedure catalog, semantic views), and serves users through an internal web interface. Inference runs on local models hosted on the same or adjacent hardware, or, where policy allows, through an approved endpoint. SyteRay implements this pattern with a hard architectural rule: no SaaS-only dependency anywhere in the stack, so every component can run inside the firewall, and air-gapped installation via offline bundle is a supported path.
What It Means for Security Review
On-prem AI turns the hardest review questions into short ones. Where does ERP data flow? Inside the network.
What third parties process it? None required. What is the egress story for prompts and answers?
There is none to tell. For CMMC assessments and ITAR postures, the data-flow diagram stays inside the boundary, and per-user accounts, role-based access, and a permanent audit trail on any data changes give assessors the controls they expect from any internal system touching the ERP.
- No required egress: questions, answers, and data stay inside the network.
- Air-gap supported: offline installation bundle, local models only.
- Per-user, role-based access with admin-gated operations.
- Writes disabled by default per deployment, with full audit and undo when enabled.
The Realistic Footprint
This does not require a GPU cluster. A single VM in the 16 GB RAM class is a sensible starting point for a working deployment, scaling up with user count and heavier local models. The install builds the knowledge layer automatically, schema discovery, the stored-procedure crawl (about 45 minutes for the reference deployment's 8,636 procedures), and semantic view creation, so the DBA-facing work is measured in hours: provision the VM, grant read access, run the installer.
From there it is an ordinary internal application: your server, your backups, your logs.
Try it here, free
Sovereign AI Readiness Assessment
Score your organization across eleven dimensions of sovereign AI readiness, from data residency and model provenance to cleared personnel and air-gapped operations.
Frequently asked questions
Can AI over ERP data pass CMMC or ITAR review?
On-premise architectures can, because the data-flow diagram stays inside the boundary: no ERP data or derived prompts leave the network. Cloud-only AI tools routinely fail exactly this question. Air-gapped operation with local models is the strictest configuration.
What hardware does on-prem AI for SyteLine need?
A single VM in the 16 GB RAM class is a realistic start, scaling with users and model size. Local LLM inference benefits from more memory or a GPU, but a working evidence-first deployment does not require cluster-scale hardware.
Does SyteRay work fully air-gapped?
Yes. An offline installation bundle and local model inference are supported, so SyteRay runs with no internet connectivity at all. This is a designed-for configuration, not a degraded mode.
Get an on-prem AI deployment plan for your SyteLine environment. Contact Netray.