// sovereign

Erebine, in your datacenter.

The platform behind erebine.ai, licensed per node to run inside your network on your own accelerators. Ask for a 30-day evaluation, or talk to the engineers who build it.

What the license includes

  • An operational license for the full stack, granted per node.
  • Container images, packages and binaries, ready to deploy.
  • Internal platform documentation and deployment instructions.
  • Direct access to engineering, the people who built the router and the inference mesh.
  • 24x7x365 operations for inference and workspace.
  • Co-development on the integrations your deployment needs.

What runs on it

  • Erebine Inference Microservices (EIMs) serving models on your accelerators, and Erebine Execution Managers (EEMs) running tool and shell calls for your workspaces.
  • An OpenAI-compatible API: chat completions, Responses, embeddings, speech, batch and realtime transcription, translation, images, files, batches and rerank.
  • Three project roles. Owners hold billing and approve destructive changes, operators run configuration and members, members use the project.
  • API keys with scopes, an expiry, a request quota and an IP allow and block list, pinned to one endpoint or a set of workspaces.
  • Approvals before tool calls run: a write waits for an owner or operator, a destructive call for an owner.
  • An audit log of admin and security actions that cannot be edited once written, and a per-tool-call trail you export as CSV or JSONL.
  • Stored content encrypted at rest with AES-256-GCM under per-project keys, and zero data retention per workspace.
  • Usage per project, spend alerts, SLO tracking, maintenance windows and a status page.

Licensed per node

Billed monthly, six-month minimum. EEMs are included. Per-node rates are quoted on request: choose Talk to engineering in the form. A frontend node runs the API and the workspace UI, a router node the scoring router and the inference mesh, and an EIM serves your models. Source code, hands-on deployment and custom integrations are available at additional cost. Term length beyond the minimum, isolation guarantees and add-on scope are set per contract. Cost Comparison sets the license beside hosted and Self-Hosted.

How can we help?

Every field is required unless marked optional.

What do you need?

Model, count and hosts, the way you would say it.

We use these details only to reply about this request.

Why run Erebine sovereign

  • The hosted service, not an edition.

    The build that runs erebine.ai is the build you license. There are no edition flags: nothing is removed, throttled or held back for the private install.

  • You decide how accelerators are split.

    One EIM can span several GPUs, or several EIMs can share a machine, one per GPU or MIG slice. EIM images ship for NVIDIA GPUs (CUDA), AMD GPUs (ROCm), Intel GPUs (XPU), AMD CPUs (ZenDNN) and general CPUs.

  • OpenAI-compatible from the first request.

    Point the SDKs, editors and agents you already use at your own base URL. Chat completions, Responses, embeddings, audio, images, files and batches answer in the shapes those clients expect.

  • Licensed per node, with EEMs included.

    Frontend nodes, router nodes and EIMs each carry a monthly license. The EEMs that run tool and shell calls come with the stack.

  • A router that shows its work.

    A warm prompt is scored on sixteen signals, from model-cache affinity to deadline risk, and the EIM that serves it is drawn by weighted random sampling. Other requests follow the endpoint's standing placement or your weights, and are scored when neither applies. The signals are documented in the router docs.

  • Data stays on your side of the wire.

    The router talks to EIMs and EEMs over a mutually authenticated CurveZMQ mesh. Stored content is encrypted under per-project keys, and a workspace can be set to retain nothing.