The System One EventUtahGo To Event Page →

Midium

AI should be owned, not rented. We help you get there.

Midium Desktop is the Operating System for Local AI. Install on any M-series Mac to start using owned AI immediately, get access to cloud models with Midium Plus, or explore our Cluster options for teams and production workloads.

Use cases

What are you trying to own?

Midium Desktop is an operating system for owned intelligence — broad enough for any path. Describe what you need, or select nodes. Pathways light up. We'll shape the offer around your mix.

Or pick nodes on the grid below.

IndustryCapabilityDeploymentHelp

Nothing selected yet

Describe a use case or pick nodes — pathways will connect them, then your options open here.

Midium Desktop

The operating system for owning your intelligence

One app turns Apple Silicon into a private AI workstation — chat, code, studio, recordings, vaults, Reader and Browser sub-agents, and local APIs. Free to start on Midium Desktop.

Locallocalhost:9100
Recent chats
Release notes draft
Pricing page review
Weekly planning
v1.10.0
Sign in to Midium CloudRun Midium on hosted compute

What's on your mind?

What is blocking our launch?
0%

Gemma 4 12B·loads when you send·runs on this Mac

Type a prompt and send it. It runs on this Mac.

Illustrative demo with sample data.

Start with the OS

Own the stack. Grow when you need to.

Midium Desktop

Chat, Code, Studio, Recordings, Reader and Browser, one personal vault — free on Apple Silicon.

Need more room?

Plus unlocks unlimited vaults ($18/mo or $168/yr). Hub licenses share vaults with your team ($200/mo for a Mac mini (64 GB memory or less) and $500/mo for a Mac Studio (more than 64 GB memory)).

Need a hand?

Ask us about setup, training, or a Hub Mac we manage and ship ready for your team.

New benchmark · Inkling 2 Mini

Inkling 2 Mini, among the fastest anywhere in the world.

Thinking Machines’ new model on Midium, on its official weights and built-in drafter: up to 188 tokens a second editing a file and 155 saving one through a tool call, on a MacBook Pro.

Measured November 21, 2026 · Apple M5 Max, 128 GB · speeds are the fastest run, with the range under it.

Editing a file
188 tok/s
95–188 with MTP drafting · 95 without
Tool call, whole file
155 tok/s
103–155 with MTP drafting · 92 without
Short coding ask
105 tok/s
98–105 · first token in 0.2s
First token, 41k prompt
~16s
~2,560 tok/s prefill