§ 01 On-prem · Overview

On-Premises AI
Air-Gapped Sovereign Deployment

Deploy sovereign AI infrastructure directly in your own facility with AGICY's on-premises deployment programme. We ship Tenstorrent Galaxy RISC-V servers to your data centre, install the complete AGIOS sovereign AI stack, and provide ongoing operational support — designed so your data stays inside your building's contracted residency boundary.

Network Isolation

Air-Gap

Your Facility

Your Site

Control

Full

Design target

99.99%

§ 02 Models · Deployment

Deployment Models

Choose the on-premises deployment model that matches your security posture and operational requirements — from fully air-gapped to managed hybrid.

Air-Gapped Deployment

Complete network isolation with zero external connectivity. Your on-premises AI infrastructure operates in a fully air-gapped environment where no data — prompts, completions, telemetry, or model weights — ever traverses an external network boundary. AGICY engineers perform on-site installation of Tenstorrent Galaxy servers, configure the AGIOS inference stack, and deliver all model weights via encrypted physical media. Updates are delivered quarterly through the same secure physical transfer process. This deployment model is designed for high-sensitivity civilian workloads and meets stringent EU secure-environment standards for AI systems. Ideal for emergency management agencies, healthcare systems, and critical civilian infrastructure operators.

Air-Gapped

Hybrid On-Premises

On-premises compute with a secure cloud management plane. Your inference workloads run entirely on local Tenstorrent Galaxy servers in your facility, while a hardened, encrypted management channel connects to AGICY's EU-sovereign control plane for monitoring, updates, and fleet orchestration. All inference data remains on-premises — only anonymised operational metrics traverse the management channel. This hybrid model provides the security benefits of on-premises deployment with the operational convenience of cloud-managed infrastructure, including automated model updates, proactive health monitoring, and remote diagnostics. Perfect for enterprises that need sovereign data processing with modern DevOps workflows.

Best of Both

Managed On-Premises

Turnkey on-premises AI with full AGICY operational support. We deploy, configure, monitor, and maintain your entire on-premises AI infrastructure as a managed service. An AGICY site reliability engineer is assigned to your deployment and provides 24/7 operational support, proactive capacity planning, and quarterly performance optimisation reviews. You get the data sovereignty of on-premises deployment without needing to build an in-house AI operations team. This model includes hardware lifecycle management, firmware updates, model refresh cycles, and comprehensive SLA guarantees. Ideal for organisations that want sovereign AI capabilities but prefer to focus their engineering resources on application development rather than infrastructure operations.

Fully Managed

§ 03 Ship · Hardware

What We Ship

A complete, production-ready on-premises AI stack — from silicon to software. Every component is sovereign and RISC-V native.

Component	Specification	Details
Compute Servers	Tenstorrent Galaxy	32× Blackhole RISC-V chips per server, 612K tok/s peak
Networking	400GbE Fabric	High-bandwidth, low-latency mesh for multi-server inference
Storage	NVMe Tier	High-IOPS storage for model weights and vector databases
Operating System	AGIOS	AGICY's sovereign AI operating system — Linux-based, RISC-V native
Inference Engine	AGICY Runtime	OpenAI-compatible API with streaming, function calling, tool use
Model Library	Pre-loaded Models	Llama 4 Maverick, Mistral Large 2, DeepSeek V3, Qwen 3.5, Gemma 3
Monitoring	Observability Stack	Prometheus, Grafana, custom AGICY dashboards — all on-premises
Power Requirements	9 kW per server	Air-cooled, standard rack-mountable (42U compatible)

§ 04 Process · Implementation

Implementation Process

From initial consultation to production deployment in 6–8 weeks. Our structured implementation process ensures a smooth transition to sovereign on-premises AI.

Phase 1: Assessment

Our solutions architects conduct a comprehensive site survey of your data centre facility, evaluating power capacity, cooling infrastructure, network topology, and physical security controls. We work with your team to define workload requirements, model selection, throughput targets, and integration points with existing systems. The assessment culminates in a detailed deployment plan with hardware specifications, network architecture diagrams, and a project timeline tailored to your operational constraints and security clearance requirements.

Week 1–2

🚀

Phase 2: Deployment

AGICY engineers arrive on-site with pre-configured Tenstorrent Galaxy servers and all necessary networking equipment. Hardware is rack-mounted, cabled, and powered on. The AGIOS operating system is installed, the inference runtime is configured, and model weights are loaded from encrypted physical media. Network integration is completed according to the approved architecture, and end-to-end functional testing validates inference performance against agreed benchmarks. Security hardening is applied per your organisation's security baseline requirements.

Week 3–5

✅

Phase 3: Production

Your on-premises AI infrastructure enters production with full monitoring, alerting, and operational runbooks in place. AGICY will provide knowledge transfer sessions for your operations team, covering system administration, troubleshooting procedures, and model management workflows. Ongoing support includes quarterly on-site visits for firmware updates and model refreshes, 24/7 remote support (for hybrid and managed deployments), and proactive capacity planning as your AI workloads grow. SLA guarantees take effect from production handover.

Week 6–8

§ 05 Questions · FAQ

Frequently Asked Questions

Common questions about on-premises sovereign AI deployment with AGICY.

What are the hardware requirements for an on-premises deployment?

Each Tenstorrent Galaxy server requires a standard 42U rack space, 9 kW of power (single-phase or three-phase), and ambient air cooling (no liquid cooling required). Minimum deployment is a single server delivering 612,000 tokens per second peak throughput. For multi-server deployments, you'll need 400GbE switch infrastructure for the inter-server mesh fabric. We support both new data centre installations and retrofit into existing facilities. Our solutions architects will assess your specific facility during the Phase 1 site survey and provide detailed power, cooling, and space requirements tailored to your deployment scale.

How does the installation process work for air-gapped environments?

For air-gapped deployments, AGICY engineers arrive on-site with all hardware, software, and model weights on encrypted physical media. No network connectivity is required at any point during installation. Servers are pre-configured in our secure facility before shipping, then rack-mounted and commissioned on-site. Model weights are loaded from encrypted NVMe drives using a secure chain-of-custody process. All installation media is destroyed on-site after deployment. Subsequent updates — including firmware patches, model refreshes, and AGIOS upgrades — are delivered quarterly via the same encrypted physical transfer process, with full audit trail documentation for your security team.

What ongoing support is included with on-premises deployments?

All on-premises deployments include a comprehensive support package. Managed deployments receive 24/7 operational support with a dedicated site reliability engineer, quarterly on-site visits for hardware maintenance and model updates, proactive health monitoring and capacity planning, and a 99.99% uptime design target (MSA-defined credits if contracted — not a present-tense guarantee). Hybrid deployments include remote monitoring through the encrypted management channel. Air-gapped deployments receive quarterly on-site support visits with pre-scheduled maintenance windows. All support tiers include access to AGICY's sovereign AI engineering team for architecture consultations and workload optimisation guidance.

Can an air-gapped deployment receive model updates?

Yes. Air-gapped deployments receive model and software updates through our secure physical media transfer process. Quarterly update packages include new model releases, optimised model weights, AGIOS patches, and inference runtime improvements — all delivered on encrypted NVMe drives with cryptographic integrity verification. Your security team validates each update package against published checksums before installation. AGICY engineers can perform updates on-site, or your trained operations team can apply updates independently using our documented procedures. For organisations requiring faster update cycles, the hybrid deployment model provides automated updates through the encrypted management channel while keeping all inference data on-premises.

§ 06 Get Started · Contact

Deploy AI On-Premises

Bring sovereign AI infrastructure into your own facility. Air-gapped, fully controlled, and built on open RISC-V silicon. Talk to our solutions architects today.

Deploy AI On-Premises →Explore Compute

compute hardware/tenstorrent trust-center solutions/sovereign-cloud