Enterprise · On-premise

On-premise voice AI on your own GPUs

KugelAudio text to speech runs in your data centre or private cloud, on your Kubernetes cluster. Speech is generated on your hardware, and your applications call the same API they would call in the cloud.

Read the deployment guide

Enterprise · On-premise

What runs in your cluster

  • Speech generated on your hardware

    The text-to-speech model, the optional text normalizer and the session store run as one Helm release on your GPU host. Requests reach the API inside your network, behind your own gateway and TLS, and the text and the audio stay there. Voices and their reference recordings are kept in the KugelAudio voice catalogue.

  • The API your code already uses

    Point the Python, TypeScript or Java SDK at your deployment’s URL. Listing voices and generating speech work as they do against the hosted API.

  • Upgrades on your schedule

    Every release is a versioned Helm package with a checksum. You install a new version when you decide to, and roll back to the previous one with Helm.

What stays connected

One outbound connection, no air gap

Your deployment keeps an outbound HTTPS connection to the KugelAudio API for the license, API-key authentication, usage reporting and voice lookups, and pulls its images from our registry. If that path is down, requests cannot be authenticated. Fully air-gapped operation is not supported.

What you need

  • Linux with a supported Kubernetes or k3s version
  • Helm 3 and kubectl, run by a Kubernetes administrator
  • An NVIDIA GPU with a working driver and container runtime
  • Outbound DNS and HTTPS to the KugelAudio API and container registries

The standard installation is a single machine with k3s and one GPU. Multi-node installations use the hardware profile we supply with the release.

How a deployment starts

  1. Agree the deployment

    We agree the license, the model, the capacity and a reviewed hardware profile in an Enterprise contract.

  2. Receive the package

    You get the chart package with its checksum, the deployment license, registry credentials and the hardware configuration.

  3. Install and test

    Your administrator installs the release with Helm and runs the bundled test for health, readiness and the model catalogue.

  4. Connect your applications

    Create a project API key in the dashboard and set your SDK’s API URL to your deployment.

Cloud or on-premise?

Our guide compares the two by data flow, infrastructure, reliability, cost and who operates what.

Read the guide

Need EU hosting rather than your own hardware?

If the requirement is that data stays in the EU, the hosted EU Sovereign Cloud is included in every plan for customers in the EU and EEA, without an Enterprise contract.

See pricing

On-premise questions

Can KugelAudio run fully air-gapped?

No. The deployment needs outbound HTTPS to the KugelAudio API for licensing, API-key authentication, usage reporting and voice lookups, and to the container registries for its images.

Which models run on premise?

Text to speech, with the model agreed for your deployment and an optional neural text normalizer. Clarity speech enhancement can also be self-hosted on Enterprise plans.

What hardware does it need?

A Linux host with an NVIDIA GPU, running Kubernetes or k3s. The hardware profile we supply sets GPU placement, memory and concurrency; adding replicas does not add GPU capacity.

Do my integrations change?

Only the URL. Your SDK takes the deployment’s address as its API URL and a project API key, and the requests stay the same.

How do upgrades and rollbacks work?

Each version arrives as its own package with a checksum. Your administrator upgrades with Helm when you choose to and can roll back to the previous version.

How is on-premise priced?

In an Enterprise contract, together with SLAs, a custom DPA and committed rates.

Plan your on-premise deployment

Tell us your hardware, your capacity and where your data has to stay. We plan the deployment with you.