Overview

What denv is, what you get when you deploy a model with it, and what is live today.

denv deploys any model on any GPU cloud or your own servers. You pick a pre-built recipe and where it runs. You get one OpenAI-compatible endpoint that is fast, observable, and repairs itself when a GPU fails.

What you get

  • A cluster. A real Kubernetes cluster with its own control plane, hosted by denv or by you. Its nodes can come from several providers and your own servers at once.
  • Sites. Each place your GPUs come from is a site: a provider account (Lambda, Nebius, Amazon EC2) or a group of servers you join with one command (Hetzner dedicated servers, on-prem, colo).
  • Recipes. A tested bundle for one model on one GPU type: engine and version, precision, engine arguments, serving profile, autoscaling targets, dashboards and default alerts, and the repair policy that keeps it serving. Each recipe is pinned by image digest and weights revision and tested on its GPU type before it enters the catalog.
  • Environments. One environment is one model endpoint. It holds the endpoint's config and state. Every config change creates a new version, and deploying, rolling back and scaling are deployments on the environment, each with a record you can read.
  • Repair. denv watches every GPU node for faults, drains a bad node within the recipe's disruption budget, moves traffic to healthy replicas, and resets, reboots or replaces the machine. Each step is recorded.

What denv does not do

denv does not sell or resell GPUs. You use your own provider accounts and your own servers, and each provider bills you directly at its own prices. denv charges for running the cluster your models serve from. See Pricing.

Where to go next

Status

denv is early. These pages describe the product as it works today and mark anything not yet live as coming soon. To see what runs today, write to anup@denv.ai.