01Self-hosted AI

Your data centre.
Your AI.

For teams who can’t — or won’t — send their data to someone else’s cloud. We deploy the full Arkintel platform inside your perimeter, so the models run next to the data and nothing ever leaves. Already running this way inside the Municipality of Delft.

arkintel · node-01
on-prem · live
YOUR BUILDINGeu jurisdiction
  • CHATchat & reasoning62%
  • INDEXknowledge & retrieval48%
  • GUARDprivacy gate18%
  • SCANdocument extract34%
  • VOICEtranscribe27%
  • ORCHcoordinator71%
speed142/s
latency38ms
data outnone
HEARTBEATNO INTERNET

02why self-hosted

Because ‘trust us’
isn’t an answer.

Self-hosting solves four real problems at once — data residency, predictable cost, model ownership and audit posture. Each one would justify the choice on its own.

Data residency is yours to decide

Air-gapped racks, your own VPC, or an EU-sovereign cloud — we meet you where your compliance team needs you to be.

No per-token surprises

Predictable infrastructure costs instead of pay-per-call billing that quietly explodes as adoption grows.

Open weights, no vendor lock-in

We deploy open-weight models (Llama, Mistral, Qwen, DeepSeek) that you keep running with or without us.

Audit-ready from day one

Full logging, permissioning and review trails designed to pass GDPR, DORA and sectoral audits without drama.

03what runs inside

The full suite,
inside your perimeter.

Not a single-purpose chatbot. Five production apps your team uses every day, plus the custom modules we build between them — all open-weight, all on your hardware. Each tile links to the app’s own page.

// also inside the boxOpenAI-compatible API for your own appsSSO · RBAC · SCIM · full audit logQuarterly model upgrades, monitoring & evalsMulti-tenant or single-tenant — your call

04inside the perimeter

What it actually
looks like, running.

A self-hosted Arkintel deployment is one self-contained node that lives inside your perimeter. Your data flows in. Models do their work on your hardware. Answers come back out of the rack — nothing else does.

// same shape on-prem, in a private cloud, or fully air-gapped — the constraints change, the picture doesn’t.

fig 01 · your arkintel nodelocation: your buildingdata leaving: none

  1. Sovereignty perimeter

    A boundary your team owns end-to-end. On-prem, in your VPC, or fully air-gapped — same software, your rules.

    the dashed ring

  2. Models on your hardware

    Open-weight models served next to your data. No license lock-in. No per-token billing. If we go away tomorrow, your AI keeps working.

    the rack in the centre

  3. Data in. Nothing out.

    Wire in the systems you already run — case files, tickets, archives, records. Answers come back inside the perimeter; nothing leaves.

    the teal pips

  4. Already shipped here

    Live inside Delft, TU Delft, JustAskMomo and Schuldhulpje — every one a private node, deployed on its owner's terms.

    the orbiting chips
// for the security & platform teamReference architecture & CLI→Compliance matrix · GDPR · NIS2 · DORA→

05how we deploy

From first call
to live platform.

Six to ten weeks from kick-off to a deployed pilot — depending on your data, your infrastructure and your compliance environment. One senior team start to finish, no handoffs, no subcontractors.

  • 30min

    first call

  • 1page

    deployment outline

  • 6–10weeks

    to production

  • 0

    data leaving

  1. step 01

    We assess

    A short engagement to understand your data, your users, your regulatory environment and the hardware you have (or need).

    30 min call · one-page outline
  2. step 02

    We deploy

    We install the full stack on your servers — models, vector DBs, retrieval, UI, auth. Air-gapped if required.

    Models · vector DB · auth · UI
  3. step 03

    We tune

    We ingest your documents, wire the integrations, fine-tune where it matters, and ship a real product to real users.

    Ingest · evals · fine-tune
  4. step 04

    We stay

    Quarterly model upgrades, monitoring, evals and training. A partnership — not a drop-off.

    Quarterly upgrades · monitoring

06live in production

Already running
inside the City of Delft.

A fully on-premises AI platform for civil servants — private chat, document search and in-house transcription, running on the city's own hardware. Live in production. No data ever leaves the building.

Municipality of Delft

DelftGPT

Municipality of Delft

We need our public servants to use AI safely — without anything ever leaving the building.

— the brief from Municipality of Delft

// what's running

ChatKnowledgeTranscribeSelf-hosted
outbound packets
0
civil-servant use
Daily
GPUs in Delft
On-prem

07Arkintel-hosted · EU cloud

No data centre? Use
ours — in Europe.

Many municipalities and small businesses don’t have a server room, or the team to operate one. So we run the same stack for you, on European-jurisdictional cloud infrastructure we manage end-to-end.

European jurisdiction

Deployed on European-headquartered, EU-resident cloud providers. Beyond the reach of the US CLOUD Act and any foreign subpoena.

Keys stay yours

Customer-managed encryption keys, dedicated tenancy, full audit log. We operate the platform; you control the cryptography.

Same stack, same models

Open-weight models. Same APIs as our self-hosted deployments. If your infrastructure situation changes, we move with you.

Operated by us

Patches, model upgrades, monitoring, evals — all handled by Arkintel’s team. You get a managed service; your data stays under European law.

eu cloud
  • jurisdictionEU · NL/EEA
  • tenancydedicated
  • egressnone to non-EU
  • operationsArkintel team

// who it’s for
Public-sector teams without their own racks. SMBs whose clients won’t accept US clouds. Anyone who needs sovereign AI now and infrastructure later.

// not sure which mode fits?

Talk through deployment options→

08common questions

What every decision-maker
asks on the first call.

Honest answers to the four objections that come up every time we sit down with a CIO, a security leader or a procurement lead. The fifth one is always different — and that’s what the first call is for.

  • Not necessarily. Modern open-weight models run comfortably on a single well-specced server. We'll size it for your load — from one box to a full rack.

09see also

Run more than just the stack.

Self-hosting answers "where". These pages answer "what runs inside it" and "who's already doing this".

app

want frontier models too?

Add Privacy Gate

Self-hosted on its own keeps everything home. Add Privacy Gate and you can route safe prompts to the best frontier model — and keep sensitive ones on your private deployment.

solution

solutions

For regulated industries

Government, healthcare, legal, finance, family offices — sovereign AI under European law.

build

in production today

DelftGPT

Fully on-prem AI, hosted inside the municipality.

page

deep dive

For your technical team

Architecture, stack, CLI, compliance matrix — for SREs, security teams and platform engineers.

ready when you are

Let's put AI inside your
own four walls.

Tell us about your data, your compliance environment and your infrastructure. We'll come back with a concrete deployment plan — usually within a day.

Start a project