SOLYSTOPIAProject Hearthmind, tools
The economics of local vs. cloud AI
Four calculators and a blueprint generator, all built on transparent formulas in a typed module. Change any assumption and the numbers move with it — nothing here is a hidden model.
Local vs. cloud, total cost of ownership
Depreciate the capital cost over its useful life, add electricity and maintenance, and compare against an equivalent monthly cloud inference bill.
Depreciation is straight-line over useful life. No resale value, financing cost, or inflation is modelled — treat this as a directional comparison, not a financial guarantee.
Renewable compute, return on investment
Solar and battery capital against the grid electricity it displaces over the system's life — the case for scheduling compute against sunlight rather than the grid.
No feed-in tariff, panel degradation, or battery replacement cost is modelled. Site-specific solar yield varies with orientation, shading, and UK regional irradiance.
Waste heat recovery, value estimate
Nearly all electrical draw from compute becomes heat. This estimates its value if captured during the heating season, discounted by a recovery efficiency you set explicitly.
No heat-recovery hardware has been built or tested for this project yet — these figures are a planning estimate, not a measured result. Recovery efficiency below 0.4 is a conservative starting assumption for a simple air-to-air or air-to-water exchanger.
Data sovereignty premium
A deliberately subjective figure: what would you pay, per query, to keep that query's data on your own hardware? Set a sensitivity weight and see what it implies against current cloud spend.
This is a values-based number, not a market price — there is no established market for household data sovereignty. It exists to make the trade-off explicit, not to imply precision.
Sovereign blueprint generator
A transparent, rule-based recommendation — not a black box. Every output traces back to the assumptions listed underneath it.
Household AI
One NVIDIA DGX Spark-class unit (≈128GB unified memory) as the household’s shared node.
Grid-powered for now. Solar and battery coupling is the natural next stage once the compute baseline is proven.
Serious open-weight models (tens of billions of parameters) for daily household use, RAG over a personal archive, and moderate concurrent workflows.
- 01Confirm 1m² is enough for the chosen hardware plus ventilation and maintenance access.
- 02Measure baseline household electricity tariff and typical daily usage window before sizing energy coupling.
- 03Acquire the household ai hardware tier and record capital cost, delivery date, and warranty terms.
- 04Install with a smart plug or equivalent power meter from day one — this is the entire measurement dashboard input.
- 05Prioritise memory capacity and context length for long-document reasoning over raw tokens/sec.
- 06Run for at least one full month before evaluating whether capability is sufficient or escalation is genuinely needed.
- 07Publish measured results (kWh/day, tokens/kWh, avoided cloud spend) as a Hearthmind experiment log entry.
- 01Hardware pricing assumed at list price in GBP; actual landed cost varies with import duty and currency movement.
- 02Expected capability is a qualitative estimate based on published unified-memory model-size guidance, not a benchmarked guarantee.
- 03Relevance score reflects fit with the Hearthmind subsidiarity framework, not a general product rating.