dl model · own GPUs

Run open models on your own GPUs, and update them safely.

Last week the harness was local. This week the model is too. dl model up checks your GPU, picks a tested config, starts the model, tests it, and connects the dl coding agent to it. Your prompts, code and weights never leave your hardware.

curl -fsSL https://dl.new | sh
dl model up

Have an Nvidia GPU with 24 GB or more? dl offers to set up a model on its first run. Needs dl 0.30.1 or later.

Read the docs →Talk to us about your setup

Why

New model versions shouldn't cost you days.

Teams that self-host for privacy already have the hardware. The hard part is running a model and moving to the next version without breaking the agents that depend on it.

dl model upChecks your Nvidia GPU, picks a tested config, starts the model with vLLM, tests it, and points the dl coding agent at it.
dl model updateStarts the new version next to the old one, tests both, and only switches if the new one passes.
dl model rollbackBrings back the previous model with the same settings.
Checks that it answers correctlyBefore a model goes live, dl checks real answers (a chat, speed, a tool call and a JSON reply), not just that the server responds.

Today

What works now, and what's next.

NowOne Nvidia GPU, one model per machine, vLLM. Setup, tested updates and rollback. Tested with small models.
Coming soonModels that need several GPUs, such as GLM 5.3 Flash and DeepSeek V4.1 Flash, several models on one machine, and SGLang.
On the roadmapAMD and Huawei Ascend. Tell us in the form below if you want early access.

dl model docs Try dl with free tokens

Design partners

Bigger setup? Talk to us.

We're working with a few teams that own GPUs and run open models. Tell us what you run, and we'll help you set it up. The same form is the early-access list for AMD and Huawei Ascend.