Stratumus
Coming late 2026 Join early access

Turn the Macs in your home into AI you can use anywhere.

Set up local models on the Macs you already own. Then reach them from your iPhone, a laptop, or any device with a browser, all from one private workspace.

Join early access See how it works

macOS firstoMLX and OllamaNo paid inference API required

Illustrative. Model names are examples. Pairing several Macs and using them from any browser works today. The macOS Connector, model discovery and choosing between Macs are in development.

Why it matters

Your home already has the hardware. Now it has one front door.

Most households have Macs that spend much of the day idle. Stratumus puts them behind a single private workspace, so any device you own can reach the Mac that has the model you want.

01 — Reach

Every Mac, from any device

Pair the Macs you own, then reach them from an iPhone, a laptop, or anything with a browser, including as an installed web app.

02 — Models

Your choice of local AI

oMLX and Ollama today, behind one chat. Switch runtimes without changing how you work.

03 — Cost

No bill for every request

Models run on hardware you already own, so you are not paying an inference provider per request.

04 — Control

Primary and fallback Macs

Choose which Mac answers first and which steps in when one is offline. Rename or unpair any Mac.

In development

Setup

From download to private AI in three steps.

This is the installation experience currently being productized for early access.

  1. 1

    Install Stratumus Connector

    Open one macOS app. No Node.js, ports, or configuration files.

  2. 2

    Pair your Mac

    Approve the connection from stratumus.com.

  3. 3

    Start chatting

    Stratumus discovers oMLX, Ollama, and compatible installed models automatically.

Hardware

Every Mac gets a model that fits.

Apple silicon shares one pool of memory between macOS and the model, so RAM decides how large a model a Mac can run comfortably. Here is a sample household.

Home Mac

Mac mini

16 GB

Comfortable up toabout 9B models

  • Qwen3.5-9B-MLX-4bit≈ 5.5 GBEveryday chat, writing and light coding
  • Llama 3.2 3B≈ 2 GBFastest answers on the lightest load

MacBook Pro

Laptop

24 GB

Comfortable up toabout 14B models

  • Qwen3 14B≈ 9 GBA balanced everyday model
  • Gemma 3 12B≈ 8 GBWriting, and reading images

Office Mac

Mac mini

32 GB

Comfortable up toabout 30B models

  • Gemma 3 27B≈ 17 GBLong documents and careful writing
  • Qwen3 30B-A3B≈ 19 GBFast reasoning and coding, with only about 3B parameters active at a time

Sample hardware. Sizes are approximate for 4-bit models, and real memory use depends on the model, quantization and context length. As a rule of thumb, keep a model under about two-thirds of a Mac's memory. Model names are examples, not a supported-model list.

How it works

The model runs on your Mac. Stratumus carries the conversation.

Your selected model runs on hardware you own. Stratumus securely relays requests between the web app and your Mac, so you can use it from anywhere. Each chat is answered by one Mac.

  1. Web app or PWA

    Where you type, on any device.

    Your deviceEncrypted connection

  2. Stratumus control plane

    Signs you in, keeps your conversations, and relays requests to your Mac.

    Hosted by StratumusAuthenticated outbound relay

  3. Stratumus Connector

    Runs on your Mac and connects out to the control plane.

    Your MacLoopback only

  4. oMLX or Ollama

    Runs the model you picked, on your hardware.

    Your Mac

What Stratumus handles

What passes through

Prompts and responses pass through the Stratumus control plane, currently hosted on Railway, on their way between your browser and your Mac. Direct or end-to-end encrypted transport is planned.

What is stored

Conversations are managed by the Stratumus application, which is how your history persists between sessions.

What changes with web search

Web search is optional. When you turn it on, your search query goes to an external search service to fetch results.

Status

What's proven, what's in progress, and what's next.

Validated in the current build
  • Secure authentication
  • Persistent conversations
  • Streaming responses
  • Optional web search
  • Local oMLX and Ollama inference
  • Multiple paired Nodes
  • Stratumus PWA
  • No required paid inference API
  • Authenticated outbound relay and reconnection
In development now
  • Downloadable macOS Connector
  • Browser-based pairing
  • Automatic provider and model discovery
  • Primary and fallback management
  • Rename and unpair controls
  • Clean installation without Terminal
Planned
  • Direct or end-to-end encrypted transport
  • Additional local engines beyond oMLX and Ollama

FAQ

Common questions.

Does Stratumus combine my Macs into one bigger AI?

Not yet. Today, each chat is answered by one Mac running one model. Turning the extra hardware you already own into a shared server for bigger models is where we're heading. It's a concept for now, not part of the current build. We're passionate about putting the compute power we already have to work as our own AI infrastructure. We like the way you think!

Do my chats leave my home?

Inference stays on your Mac. Prompts and responses pass through the Stratumus relay so you can reach your Mac from anywhere. See exactly what Stratumus handles.

What do I need?

A Mac running oMLX or Ollama, and a browser on the device you are using. Your Mac connects out to Stratumus, so there are no ports to open, and it works away from home. Stratumus is macOS first.

The Stratumus mark

Get early access.

Stratumus is starting on macOS with oMLX and Ollama. Leave your email and we'll let you know when early access opens.