03 / In Practice
Why on-prem inference matters.
Cloud-based AI assistants leak proprietary procedures, asset configurations, and security details to model providers. On-prem inference keeps your operational knowledge inside your network.
Your senior technician has 20 years of building knowledge. He retires next year. Virtual Technician trains on your runbooks, work-order history, and asset records to answer the questions your junior staff would otherwise have to interrupt him for.
Four capabilities across training, retrieval, and verification.
PDF, Word, image, and CAD runbooks ingested and indexed. SOPs, manufacturer manuals, and tribal-knowledge docs.
Every answer cites the document, page, and section it came from. No hallucinated procedures.
LLM runs on your infrastructure. No data leaves your network. Air-gap compatible.
Technicians upvote/downvote answers. Subject-matter experts approve canonical responses. The model improves continuously.
Cloud-based AI assistants leak proprietary procedures, asset configurations, and security details to model providers. On-prem inference keeps your operational knowledge inside your network.
On-premise by default. Cloud-deployable when required. Your facility data never leaves your network.
TLS everywhere, secrets in vault, row-level security on every query. Containerized and isolated so one service going down never takes the rest with it.
A·IQ is built by Arcis FM, a Service-Disabled Veteran-Owned Small Business (SDVOSB). Set-aside eligible for federal contracts. CAGE 14DG6 · UEI Z95MQL2KEYG3.
Engineer and operator questions on this capability.
Configurable. Default is a quantized open-weights model sized for on-prem GPU. Air-gap deployments use the smaller variant.
Yes. Retrieval-augmented over the asset registry, work-order history, and runbook library.
Every answer cites sources. Technicians flag bad answers; SMEs approve canonical responses; the system improves.
A modest GPU (16-24GB VRAM) for typical sites. Larger portfolios scale horizontally.
Each Nexus module is sharper paired with the next.
Send us 10-20 of your most-used runbooks. We will ingest them and demo the assistant against real questions your team asks.
