SOLYSTOPIA
Open correspondence
Build / The Sovereign AI Harness

Less dependency.
More capability.

Understand a system before you abstract it. The tools here are not here to make you dependent on another dashboard — they help you see what you already have, build what you can own, and share the method.

Lane 01 / Diagnose

See where the household is exposed.

The Household Capability Audit maps practical strengths, gaps, dependencies, and the next three useful actions.

Run the Household Audit
Lane 02 / Build

Make one owned capability real.

Start with local intelligence: a small model, a computer you control, and a documented setup you can explain to someone else.

Capability 01 is live below

The Sovereign AI Harness

Ten stages, understood before they are abstracted.

Turn a local model into useful local intelligence — one capability at a time. Each stage below is a real building block, not a marketing category: two are live and working today, the rest are named honestly as planned.

  1. 00

    Model

    Live

    Pick a small, open-weight model you can hold in your hand — a GGUF file with a known provenance, not a hosted API key.

    Try it below →
  2. 01

    Runtime

    Live

    The runner that loads the model and lets it think — llama.cpp today: no account, no hosted dashboard, no hidden subscription.

    Try it below →
  3. 02

    Interface

    Planned

    How a household actually talks to the model — a terminal today, a proper local chat interface next, voice and household devices after that.

  4. 03

    Memory

    Planned

    Giving the model context that persists — notes, documents, and conversation history that stay on your machine, not a vendor server.

  5. 04

    Tools & actions

    Planned

    Letting the model act, carefully — read a file, check a calendar, draft a message — each capability granted deliberately, never by default.

  6. 05

    Orchestration

    Planned

    Running more than one model or agent and routing tasks between them, so the household is not betting everything on a single point of failure.

  7. 06

    Data ingestion

    Planned

    Feeding a household’s own documents, manuals, and archives into a system it controls — the job Hearthvault is designed to do.

    See Hearthvault →
  8. 07

    Networking

    Planned

    Reaching the system from other devices in the house without exposing it to the open internet — the job Hearthnet is designed to do.

    See Hearthnet →
  9. 08

    Power

    Planned

    Coupling compute to real household energy — solar, battery, and a scheduler that routes heavy inference onto surplus generation.

    See Hearthpower →
  10. 09

    Physical systems

    Planned

    The dedicated machine itself: a DGX Spark-class node, measured on cost, power, space, and repairability — Project Hearthmind’s current focus.

    See Project Hearthmind →
Capability 01 / Local Intelligence13% complete
01 / Understand

What “local” actually means

Local intelligence means the model runs on hardware you control. Your prompts, notes, and documents do not need to leave your home network.

It is not magic, and it is not a replacement for every cloud model. It is a smaller, inspectable capability with a different bargain: more ownership, less convenience.

Honest boundaryA 0.8B model will be slower and less capable than ChatGPT on many tasks. That is acceptable here: the first goal is to understand the system and keep the option of private, offline use.

Your place is saved on this device. No account required.

What comes next

One capability at a time.

Future stages will cover memory, tools and actions, orchestration, and the physical systems above — each published with its limits, ownership model, and a way to leave.

See Infrastructure 001: The Ward AI CentreSupport the Founding Network