Blog · notes from the workshop
What we learned building an AI you actually own
Start with what it is. Then the week that made it timely. Then one mechanism taken apart, down to the bytes. Below those, the guides: real hardware, real engines, every command checked. Each post stands on its own and names its sources.
Launch week · August 25, 2026 · three parts, in reading order
-
Part 1 · Start here · seven minutes
Introducing SovereignAI
One AI that runs on your machine, thinks with models you choose, remembers with receipts, and proves you can leave. What it is, why it exists, the twelve things you own when you run it, the three questions we answer about ourselves, what is not sovereign yet, and the two-minute trial.
Read it → -
Part 2 · The week · nine minutes
Local is solved. Ownership isn't.
In one August week, Perplexity and NVIDIA shipped a local-first agent, a Berkeley lab shipped an engine that runs 120B-class models on a gaming GPU, two of the year's best open models dropped their weights, and two benchmark sites launched to help you choose. What that week means for sovereign AI — and what none of it says about who owns the result.
Read it → -
Part 3 · Deep dive · ten minutes
The customs declaration, in depth
How SovereignAI shows you the exact bytes before they leave your machine: the side-effect-free preview, the manifest, the receipt on every remote answer, the parity test that proves the preview is the request, and the two things we deliberately refused to build.
Read it →
Guides · setting it up on real hardware, with real engines
-
Guide · twelve minutes
SovereignAI on NVIDIA: DGX Spark, RTX desktops, RTX workstations
Which install path fits each NVIDIA machine, how to give Ollama the GPU, which models genuinely fit at 8, 24, 48, 96 and 128 GB, when FreeToken applies — and what does not work yet, including the Spark’s ARM CPU.
Read it → -
Guide · ten minutes
FreeToken in SovereignAI: frontier-class sparse models on the GPU you have
Requirements, install and serve, enabling it in Settings, what the doctor and the starter shelf tell you, sizing per RAM, the Docker wiring, the reasoning panel, and the limits.
Read it → -
Guide · eight minutes
Liquid AI’s device-native LFM models in SovereignAI
Pull LFM2.5 through Ollama, give a small Liquid model the cognition role so it writes your memory while anything else chats, build a persona on it, and read the license first.
Read it → -
Guide · eight minutes
SovereignAI with Perplexity’s Portable Computer: what works today
Same machine, shared GPU, two different jobs — with a hard line between what we verified, what Perplexity has stated, and what nobody has documented yet.
Read it →
Signals · what shipped elsewhere, weekly, assembled automatically
Each week a machine reads the release feeds and model registries listed in the repository and formats what is new. No summarizing, no commentary, no editor — every line is a project's own announcement, linked back to it. The written posts above are the ones we stand behind; these are the raw record beside them.
- Week of August 30, 2026 20 items
House rules. Every post links its sources. Every product claim names the shipped mechanism behind it, and where something is not sovereign yet we say so — the same rule as the Sovereignty Ledger. No newsletter, and no tracking at all — the counter this used to point at was never switched on, so it was removed rather than left standing as a claim.
Following along. There is no mailing list to join, because a mailing list is a copy of your address in our hands. Instead there is an Atom feed: paste that link into any reader and it pulls new posts on your schedule, tells us nothing, and needs no account. Reading with the feed is the sovereign way to read a blog, and it would be strange for this one not to offer it.