Midium Desktop
Local AI on your Mac. Free.
An engineer gets a local inference engine, a personal Vault, and MCP sub-agents for the editor they already use. Speed up the work, or lower the AI coding bill. Local inference and your personal Vault stay on your machine by default. Connectors only send what you point them at.
Free. macOS on Apple Silicon.
Local inference
Host models on your machine and experiment with them there. Tool calling, speed, and both flex and static APIs, tuned for Apple Silicon.
Personal Vault
One context Vault on the machine. The context stays on your machine.
MCP sub-agents
A code harness built for sub-agents. MCP works in Cursor, Codex, and Claude as soon as you plug it in.
48 GB of unified memory
The app runs on Apple Silicon. 48 GB of unified memory is recommended. That covers local models, the Vault, and sub-agents at the same time. The app checks what fits before it starts a model.
Commercial
Dedicated capacity, or a Hub license.
Commercial use adds tunneling, remote access, shared Vaults, and production inference. Get that on dedicated capacity we run, or with a Hub license for hardware you run. A Hub license is $200/mo for Starter and Standard, and $500/mo for Pro, Max and Ultra.
- Auto web API tunneling, so a local engine can serve a product.
- Private Relay: end-to-end encrypted access to your hub, scoped to your team. No public URL, no port forwarding, and Midium's servers never see request contents.
- Unlimited Vaults, and team members with scoped permissions.
- Production inference capacity shared by the team.