MUModel Uplink

OUTBOUND ONLY · END-TO-END TLS · OPENAI-COMPATIBLE

Your model.
One secure URL.

Model Uplink gives the Ollama or vLLM server already running on your hardware a permanent, authenticated URL. One command. No port forwarding, no VPN, and encryption that ends on your machine, so not even we can read a prompt.

Connect your modelSee the wiring

Works with the OpenAI SDKs you already have. macOS and Linux. Runs as a launchd or systemd service.

home-gpu — zsh
$ modeluplink serve ollama --name home-gpu✓ ollama 0.30.8 found at 127.0.0.1:11434✓ linked to eu.relay.modeluplink.com outbound only✓ certificate issued private key stays on this machine✓ endpoint  https://home-gpu.modeluplink.com/v1✓ key       mup_7Qx4…c9  shown once — copy it now● online · installed as a background service$
§ 01

The wiring

Your machine dials out. The relay answers to the world and forwards bytes it cannot read. Nothing ever dials in.

  1. 01

    Sign in

    A passwordless email link signs in both the CLI and the dashboard. There is no password to leak and no shared inference credential.

    modeluplink login
  2. 02

    Serve

    Point the agent at Ollama or an existing vLLM server. It registers a globally unique slug, obtains a certificate whose private key never leaves your machine, picks the closest relay region, and installs itself as a background service.

    modeluplink serve ollama --name home-gpu
  3. 03

    Point your tools at it

    Keep the SDK you already use. Change the base URL, paste a key, and the model names come straight from your local server.

    https://home-gpu.modeluplink.com/v1
§ 02

One model, every place you work

The same weights, the same address, from the editor to the agent runner to the product you ship.

EDITORS

Code with your own model

Chat, edits, and completions in Continue, Cline, and any extension that takes an OpenAI base URL.

VS Code · JetBrains · terminal
AGENTS

Run agents anywhere

Automations keep working from a laptop, a CI runner, or a cheap VPS while the GPU stays under your desk.

OpenAI SDK · agent frameworks
CHAT + RAG

Keep knowledge private

Open WebUI, document search, and internal assistants that never send a document to a third-party model.

chat UI · embeddings · RAG
YOUR PRODUCT

Ship on hardware you own

Prototypes, SaaS features, bots, and mobile apps backed by a card you already paid for.

one URL · scoped keys · no port

Tool use, vision, and response-format support depend on the model and engine you run locally.

§ 03

Two lines change. Nothing else does.

Every OpenAI-compatible client works unchanged, from a Python script to the extension in your editor. Swap the base URL, hand it an endpoint key, and ask for whatever model Ollama or vLLM has loaded.

Get your URL
python
from openai import OpenAI

client = OpenAI(
  base_url="https://home-gpu.modeluplink.com/v1",
  api_key="mup_…",
)

client.chat.completions.create(
  model="qwen3:32b",
  messages=[{"role": "user", "content": "hi"}],
  stream=True,
)
vs code · continue · ~/.continue/config.yaml
models:
  - name: home-gpu
    provider: openai
    model: qwen3:32b
    apiBase: https://home-gpu.modeluplink.com/v1
    apiKey: mup_…
curl
curl https://home-gpu.modeluplink.com/v1/models \
  -H "Authorization: Bearer mup_…"
§ 04 · WHAT WE CAN SEE

Your prompts stay between you and your computer.

Everything an app sends to your model is encrypted on the way in, and the only key that can open it lives on your own machine. Our servers pass that traffic along the way a courier delivers a sealed envelope: they know where it is going and how heavy it is, but they cannot open it. We can see that your endpoint was used, from which region, how much data moved, and when. We cannot see what was said.

One thing worth knowing: if your computer is off or offline, apps will get a plain “could not connect” error rather than a friendly message from us, because nothing on our side can answer for your machine.

Questions about the security model → contact@modeluplink.com

The GPU under your desk, reachable from anywhere.

$ modeluplink serve ollama
Connect your modelRead the terms

One subscription, billed per active endpoint. Cancel from the billing portal any time.