Vela Anvil turns your Vela Context Data Hub into something you can simply ask. It is software, delivered as two components: the Anvil Gate deploys on your hub and decides what any AI may see - per role, per client - serving scoped retrieval from your live data, knowledge graph, and history, with a complete audit trail. The Anvil Runtime deploys on a GPU node you own and runs European open-weight models that answer questions grounded in your operation, with lineage back to the exact entities, events, and documents behind every answer.

No cloud. No data leaving your network. Air-gap capable. Every prompt policied, every answer logged.

Built on the Vela Context Data Hub and the ETSI NGSI-LD open standard. Designed for manufacturers, utilities, cities, and infrastructure operators that need AI on their operational data without sending a single byte offsite.

AI IS THE HAMMER. VELA ANVIL IS WHAT IT STRIKES ON.

You ask: "Why did line 3 stop on Tuesday?"

Anvil answers: Line 3 stopped at 14:23 after the extruder drive tripped on overtemperature - bearing temperature rose from 68 °C to 91 °C over the prior 40 minutes, consistent with the lubrication interval being exceeded (last service logged 11 weeks ago against an 8-week schedule). A similar pattern preceded the stop on 14 March.

Grounded in your sources: SCADA temperature tags · the trip event from your historian · maintenance log entry · the machine manual · a similar past incident - linked in one graph, every claim traceable.

You ask: "Which intersections had the worst air quality during last week's roadworks?"

Anvil answers: Three intersections on the eastern corridor exceeded NO₂ thresholds during the works window, peaking near the Station Road diversion - correlated with the traffic re-routing in the works permit and a 40% rise in measured flow on the detour.

Grounded in: air-quality sensors · traffic counts · the roadworks permit · the GIS road network - linked in one graph, every claim traceable.

  FROM HUB TO GROUNDED ANSWERS IN THREE WEEKS.

PRICING

Onboarding scope: Gate deployment, AI exposure policy, Runtime installation, grounding, and acceptance testing - up to 8 days; beyond scope at day rate.

Certified hardware: NVIDIA DGX Spark, ASUS Ascent GX10, Dell Pro Max GB10, Lenovo PGX (GB10-class, 128 GB unified). Buy from any vendor, or through us at cost-plus with one invoice.

Worked example: hub Professional €1,700 + Anvil €1,490 = €3,190/month all-in for a greenfield site. Existing hub customers simply add €1,490/month.

Prices ex VAT.

INTEGRATIONS

   CONNECTS TO YOUR SYSTEMS

Your systems connect through the Vela Context Data Hub — Anvil grounds every answer on everything the hub ingests. From the shop floor to the city street.

   THE VELA LAYER UNDERNEATH 

User identity flows through to AI policy and audit. 

Start with the Sovereign AI Pilot from our Industry Starter Packs: 60 days, one certified GPU node included and yours to keep, two sources, your ten questions answered by your own on-premise AI. €9,900 fixed - fully credited if you convert to Vela Anvil. New to Vela Context entirely? The hub comes first - any Industry Starter Pack gets your data publishing, then Anvil adds the AI.

FREQUENTLY ASKED QUESTIONS?

DO WE HAVE TO BUY HARDWARE FROM YOU? 

No. Anvil is software. Run it on any certified GB10-class node - from your usual IT supplier, or through us if you want a single invoice and one accountable vendor.

WE'RE NOT AI EXPERTS.

You don't need to be. Onboarding includes model deployment, grounding, acceptance testing against questions you define, and operator training. The subscription keeps models and software updated.

ARE WE LOCKED IN?

The opposite, by architecture. The Anvil Gate is the hub's standard AI socket - any AI client can plug into it, including your own. Replace the Anvil Runtime tomorrow; the Gate keeps governing, and your data never moves.

WHY NOT JUST USE A CLOUD AI API?

Because your production data, recipes, and failure history would leave your jurisdiction and your control. For NIS2-scoped operators, that is not a price question. Anvil runs air-gapped if you want it to.

CAN IT TRAIN MODELS ON OUR DATA?

No - and we say so plainly. Anvil is built for inference, retrieval, and knowledge graphs. Grounding a strong European model in your live operational context delivers what most teams imagine training would - without an ML team or model-drift liability.

WHICH MODELS DOES IT RUN?

European open-weight models by default - Mistral (Apache 2.0), EuroLLM / Teuken, Jina embeddings. Any open-weight model on request: it's your hardware.

WHAT HARDWARE DO WE NEED?

One certified GB10-class node per AI node (128 GB unified memory, from ~€4,000). Desk-size, ~240 W, normal office or rack placement.

DOES IT REPLACE OUR DASHBOARDS OR OUR VENDORS' AI?

No. Anvil sits beside your BI, your partners' digital twins, and specialist AI models - all consuming the same governed hub. It answers questions; it doesn't replace applications.

WE DON'T HAVE A DATA PLATFORM YET — CAN YOU STILL READ OUR EXISTING SYSTEMS?

Yes. Connecting your sources is part of the deployment - OT, IoT, SCADA, MES, ERP, GIS, sensors and documents, mapped to open standards as your AI's grounding. You don't buy a platform first and the AI later; we connect your systems and stand up your AI in the same engagement.

WHICH SYSTEMS AND PROTOCOLS DO YOU SUPPORT?

OT & buildings: OPC-UA, Modbus, BACnet, MQTT / Sparkplug B (Profinet / Profibus via gateway).

IoT: LoRaWAN, NB-IoT, MQTT.

Plant: SCADA & historians (AVEVA / Wonderware, Ignition, OSIsoft PI), MES, LIMS.

Enterprise & estate: ERP (SAP, Dynamics), CMMS / EAM.

City & spatial: GIS (Esri / ArcGIS), GTFS / GTFS-RT, CKAN / DCAT-AP, open-data portals.

Plus REST, CSV, Excel and document stores. If it has an interface, it maps to open standards.

DOES IT WORK AIR-GAPPED ON THE OT NETWORK?

Yes. Anvil runs with zero outbound traffic and can sit fully inside your OT segment or secure network. Your data never leaves the building.

READY?