On-premise AI

AI that runs on your servers. Nothing leaves your network.

Knowledge assistants, voice agents and automation, with the models, vector index, connectors and audit logs all running in your own data centre. Operated by your team or by us.

Runs on
Your servers and GPUs
Data
Never leaves your network
Pricing
One-off implementation, unlimited users
You own
Source code and model weights

In short

On-premise AI means the whole AI system, including the language models, document index, connectors, voice and logs, runs on servers inside your own data centre rather than with an outside provider. DeepVox builds and deploys it on your hardware for a one-off implementation cost with unlimited users, and hands over the source code and any fine-tuned model weights.

Video spot · only visible while running npm run dev On-premise AI, explained Add a video here by saving it as: public/videos/platform/deployment/on-premise.mp4 Then run python scripts/video-posters.py for the cover image.

Why on-premise

When your data can’t leave the building.

Strict sovereignty

Professional secrecy, classified material or client confidentiality that rules out any outside processor.

Your security perimeter

The system sits behind your firewall, under your identity provider, logging and monitoring.

Unlimited users

A one-off implementation cost instead of per-seat fees, so rolling out to everyone costs nothing extra.

Yours to keep

Source code from the first commit and fine-tuned weights, with runbooks your team can operate.

What runs on your servers

Every layer, inside your network.

LayerTypical components
Language modelsOpen-weight models served with vLLM, SGLang or Ollama
Model gatewayLiteLLM, with usage and cost per department
Index and vector storePostgreSQL with pgvector, Qdrant or OpenSearch
Document ingestionDocling, Unstructured, Apache Tika, Tesseract for scans
Identity and permissionsKeycloak, Open Policy Agent, Entra ID integration
VoiceOpen speech recognition and text-to-speech
Observability and auditLangfuse, Phoenix, Grafana stack

The mapping step decides the exact components against your requirements and existing infrastructure.

How we deploy

From discovery to a running system.

  1. Discovery and sizing

    We agree the use cases, sources and users, and size the hardware: models, GPUs and storage.

  2. Proof of concept

    A working assistant on your own content, measured against questions from your team.

  3. Production build

    Deployed on your servers with Kubernetes and Helm, connected to your identity provider and sources.

  4. Handover or managed support

    Runbooks and training for your team, or managed support with a 99.9% uptime SLA.

Is it right for you

On-premise, or start in our cloud?

On-premise fits when

  • Regulation or contracts forbid outside processing
  • You already run a data centre and GPUs, or plan to
  • You want unlimited users without per-seat fees
  • You need full ownership of code and weights

DeepVox cloud may fit better when

  • You want to start in days, not weeks
  • You don’t want to buy or run GPUs yet
  • A dedicated environment in your region meets your rules

FAQ

Questions, answered.

What hardware does on-premise AI need?

It depends on the models, number of users and whether you run voice. We size it in discovery. Fine-tuned small models keep the GPU footprint modest compared with large general models.

Does any data go to OpenAI or other providers?

No. The models run on your servers. An outside provider is only used if you choose to switch one on for a specific task.

Can we start in the cloud and move on-premise later?

Yes. Start in DeepVox cloud and move to your own servers later. The platform, connectors and configuration are the same, so nothing is rebuilt.

Who runs and updates it?

Your team, using our runbooks and training, or us under managed support with a 99.9% uptime SLA.

How much does on-premise AI cost?

A one-off implementation cost with unlimited users, plus your infrastructure at cost. We quote after a discovery sprint.