The token you don't send.
Work that stays on hardware you already own does not spin up a data center on your behalf.
Same work, shorter trip
A local model answers on the machine in front of you. Nothing round-trips to a metered hyperscale cloud, and no new capacity is provisioned to serve you. The saving is proportion: the work you keep local is work the grid never routes to a warehouse of GPUs.
Reuse before rent
OpenShore runs on a computer you already have. A laptop or a desktop you keep on, awake while you work. There is no appliance to buy and no fleet to keep warm. You harness the hardware you own instead of renting someone else's.
- Your machine. Local inference uses the runtime you installed, on power you already pay for.
- Cloud on purpose. A frontier model is one deliberate tap on your own key, not the default for every token.
Measured, not hand-waved
Coming in a later release: Stack Health estimates your footprint on your device and refreshes it once a day, with energy, water, and carbon avoided versus a data center, each marked as an estimate. Nothing leaves the machine to compute it. It is not in this release yet.
The honest limits
Local inference still uses power, and a large model on a big GPU is not free. This is about choice and proportion, not a zero-carbon claim. Long runs need the computer awake, and the figures are directional estimates, not a meter reading.
Stack Health arrives in a later release, computed on your device.