Skip to main content

Compute model

Midcore is designed local-first: by default any compute that can run on your machine, does. Cloud compute is metered, transparent, and only consumed when you opt in or when an operation has no local alternative.

Principles

  1. Your machine first. Any operation with a viable local runner runs locally by default. You only pay for cloud when local can’t.
  2. Pre-flight transparency. Every cloud operation shows you its estimated cost before it runs. No silent debits.
  3. Signed receipts. Every cloud charge produces a hash-chained, signed receipt recorded on your project audit ledger. You can verify any charge offline.
  4. Sensitive stays local. Operations that handle data you wouldn’t want to ship off-device (passwords, biometric, raw camera feed) are forced local regardless of your preference.
  5. Per-tenant capacity. The local-first default isn’t a paywall trick. It’s the cheapest and the most private path, and it’s the same path our team uses.

Pages in this section

  • Local vs cloud – how the Compute Router decides per operation.
  • Pricing – what cloud compute costs, by category and by operation.
  • Security – ASAR integrity, signed WASM, sandboxed workers, and why this matters when shipping IP to client machines.

Where to set your dial

Open the Midcore Shell, choose Account on the activity rail, then Compute. Three regions you’ll touch: the routing dial (Local-first / Cloud-first / Ask each time), the per-operation overrides for fine-grained control, and the auto-approve cap that controls how much cloud compute happens without prompting.

What this changes vs other platforms

Most AI / robotics platforms charge you for every operation regardless of where it could have run. Midcore inverts that: you pay only for cloud compute, you see the price before you commit, and the platform actively prefers your hardware.

For a single user this saves a meaningful percentage of monthly spend. For a team running fifty active developers, it changes the cost curve from linear-in-headcount to roughly flat — the cloud-only operations (multi-tenant DB writes, external paid LLM API calls, large fine-tunes) cost the same per call whether you have one user or fifty, and everything else is free.