Apple's on-device Foundation Models run as a system service on every modern Apple device. There are no API keys, no per-request charges, and no usage caps. Spatial Agents leverages this for unlimited intelligence at zero marginal cost.
Cloud-based AI applications pay per token. A typical intelligence product might cost $0.01-0.10 per API call. At scale — continuous monitoring across thousands of H3 cells — this adds up to significant operational expense.
Spatial Agents sidesteps this entirely. The Foundation Model ships with the operating system. Inference runs on the device's Neural Engine. There is no metered API between the application and the model.
Generate situation reports for every active H3 cell on every tile rebuild cycle. No need to ration API calls or batch requests.
Run counterfactual interventions on every root cause in every causal graph. Explore "what if" scenarios freely without worrying about cost.
Generate reports at every resolution level — from regional overviews at res 3 to berth-level detail at res 7 — simultaneously.
| Component | Spatial Agents | Cloud AI Alternative |
|---|---|---|
| Model inference | $0 (on-device) | $0.01-0.10 per call |
| Data feeds | Free (public APIs) | Free (public APIs) |
| Server hosting | Your Mac (already owned) | $50-200/mo cloud VM |
| Scaling cost | $0 per additional device | Linear with usage |
On-device models are smaller and less capable than cloud models like GPT-4 or Claude. The context window is 4,096 tokens rather than 128K+. Spatial Agents works within these constraints through careful token budget management, payload compression, and structured output schemas that focus the model on specific, well-scoped tasks.
For the domain-specific intelligence products Spatial Agents generates — structured situation reports, anomaly classifications, causal chain narration — the on-device model is more than sufficient.