AI services · on your own infrastructure

AI on your
servers.

We deploy and maintain your company’s AI on your own servers, from the GPU to the agent.

01Your data stays put — everything runs in your data centre.
02You choose the model — and change it without touching your applications.
03Support for the whole stack — Canonical or CIQ below, us above.
Tell us about your projectSee what we do ↓
Technical team available · Mon–Fri 8:30–16:30 GMT+1
AI stack
OllamavLLMllm-dKServeKubernetesKueueMCPLangGraphDoclingCephUbuntuRocky Linux
Services

From the GPU
to the agent.

Start with any piece: a model you want to serve, a process for an agent or a server you are about to buy. Whatever is missing underneath, we build. If you are still deciding, we have guides on vLLM and Ollama and on local MCP servers.

01Serving modelsFrom the model you are trying with Ollama to serving it to the whole company with vLLM and llm-d. We measure it, deploy it and support it.ollama · vllm · llm-d · kserve→02Agents and MCPAgents connected to your systems through MCP, with narrow permissions and a log of what they do.mcp · a2a · langgraph→03GPUs on KubernetesThe same GPUs for several teams, with quotas and priorities: GPU Operator, DRA and Kueue.gpu operator · dra · kueue · mig→04RAG with your documentsAI that answers from your documents without them leaving the building.docling · pgvector · opensearch · ceph→05AI serversLenovo, Supermicro or IBM Power, chosen for the model and for how many people use it.lenovo · supermicro · ibm power→06Security and complianceAgent audits against the OWASP Top 10 for LLMs and the technical documentation the AI Act asks for.owasp · ai act · iso 42001→07Role-based trainingFive paths, from infrastructure to AI governance, at your offices or remote.5 paths · in-company · tailored→
Reference architecture

What AI
looks like in-house.

Six layers, all yours. Change one without touching the rest: Llama today, Qwen tomorrow.

L6 · Agents
Agents and MCPMCP · A2A · LangGraph · Langfuse
L5 · Models
Serving modelsOllama · vLLM · llm-d · KServe
L4 · Data
Data and RAGDocling · pgvector · OpenSearch · Ceph
L3 · Cluster
GPUs on KubernetesGPU Operator · DRA · Kueue
L2 · System
Operating systemUbuntu · Rocky Linux RLC Pro AI
L1 · Infra
GPU serversLenovo · Supermicro · IBM Power
We measureBefore you buy, we serve your model and measure it with your workload. That tells us how many GPUs you need.
We deployServer, operating system, Kubernetes, models and agents, running with your data.
We maintainOur own support for vLLM, agents and RAG, Canonical or CIQ for the operating system, and training paths so your team can run it.
Why SIXE

AI under the rules
of critical infrastructure.

We come from critical systems: IBM Power, Ceph, Linux in production. We ask the same of AI: that it runs in-house, keeps a record and that someone answers when it fails.

Open software, no lock-inOpen models and upstream software: you change model, server or vendor without rebuilding anything. If a commercial model or IBM watsonx fits better, we integrate it.
On your serversModels and data stay in your data centre, on your Kubernetes and your Ceph storage. And if your servers are IBM Power, vLLM runs there too, even without a GPU.
Support for the whole stackWe are Canonical and CIQ partners for the operating system, and we support vLLM, agents and RAG ourselves. Your team learns to run it with our role-based training paths.
OllamavLLMllm-dKServeKubernetesGPU OperatorKueueMCPA2ALangGraphLangfuseOpenTelemetryDoclingpgvectorOpenSearchCephCharmed KubeflowUbuntuRocky LinuxIBM PowerIBM watsonx
Servers

Hardware
for your AI projects.

Supermicro and Lenovo servers configured to train and serve models in your data centre. We install them for you, with the AI stack up and running. The calculations for choosing one are in our guide AI server: how to choose one. If you are looking at the RTX PRO 6000, here are its versions and which servers take it.

SupermicroSupermicro SYS-822GS-NB3RT-01-G2SYS-822GS-NB3RT-01-G2HGX B300 8-GPU · 2 TB SupermicroSupermicro SYS-422GA-NRT-02-G2SYS-422GA-NRT-02-G28x RTX PRO 6000 · 1.5 TB SupermicroSupermicro ARS-511GD-NB-LCC-01-G2ARS-511GD-NB-LCC-01-G2Grace + B300 · 748 GB LenovoLenovo ThinkSystem SR650a V4ThinkSystem SR650a V4Up to 4 RTX PRO 6000 · Intel Xeon 6 LenovoLenovo ThinkSystem SR675 V3ThinkSystem SR675 V3Up to 8 PCIe GPUs · AMD EPYC LenovoLenovo ThinkSystem SR680a V4ThinkSystem SR680a V4NVIDIA HGX B300 · 8 GPUs
Supermicro AI serversAll 29 Gold Series configurationsLenovo AI servers
You may also be interested in

Services
that connect with AI.

Kubernetes and OpenShiftThe foundation for serving models
→
Rocky Linux for AIRLC Pro AI with CIQ support
→
IBM Power11AI next to Db2, AIX and IBM i
→
IBM Storage Scale (GPFS)Data for training and serving
→
AI process automationProcesses, systems and agents
→
OpenStack as a private cloudMigrating from VMware
→
IT/OT cybersecurityProtecting data and inference
→
Sustainable technologyGetting the most out of every GPU
→
Server configuratorChoose the machine for your model
→
Contact

Let’s talk.

Mon–Fri 8:30–16:30 GMT+1.
Urgent support 24/7.

Tell us what you want the AI to do, with which data and on which servers, if you already have them. An engineer will reply with a proposal: what to build, in what order and who looks after each part.

Contact formWhatsApp
Technical team · Mon–Fri 8:30–16:30
01Emailcontact@sixe.eu→02Phone Europe+34 91 198 02 43→03Partner programmeConsultancies · MSPs · Integrators→04About SIXETeam · Track record · Clients→
SIXE