Philosophy

The engineering-log thesis

Most AI writing is either marketing copy from companies selling something or speculation from people who haven't run the thing. The space between those two is occupied by a few people who actually operate machines and are willing to publish what happens. That space is the position this blog tries to hold.

Every article comes from a real system, a real error, or a real fix. Nothing is hypothetical. If you read about an SGLang OOM here, somebody on a desk in Europe watched their model crash that way and reproduced it twice before writing it down. If you read about an upstream PR, the patch is in the commit log and the merged date is on the upstream page.

The voice is the operator's voice: first-person, technical, allergic to hype. The metaphor is the engineering log: dated entries, in-frame mistakes, no retroactive cleanup. The aesthetic is monospaced where it earns it, plain prose where it doesn't.

The sovereignty essays

Below the dated war stories runs a slower argument. This is a numbered series of essays, each one taking a single serious thinker, reading them at the source, and pressing their idea against one lived fact of running a model on this desk. Every essay stands on a receipt from this site's own log, and every essay steelmans the strongest objection before it answers it. There is no evangelism here. The case for owning your own model is weakest on cost (renting keeps getting cheaper) and weakest on frontier capability (the largest models stay in datacenters you do not control), and strongest on control (privacy, deplatforming immunity, the right to keep operating when a vendor changes its mind). The series concedes the weak ground and builds on the strong. That concession is the argument, not a hedge.

1. 01 The Radical Monopoly of Convenience Ivan Illich had a word for a tool that manufactures the need it then meters. In 2026 the rented AI API became the cleanest example I have ever seen, and the bill is arriving. Essay one of a series on sovereignty and the philosophy of running your own model.
2. 02 I Moved the Dependency, I Didn't Remove It The strongest objection to self-hosting is that it is theatre: you still sit on CUDA, on weights handed down by an oligopoly, on a rented edge. It is correct, and it is not the end of the argument. Essay two of a series on sovereignty.
3. 03 My Spark Idles at 22% and That's the Point The utilization math that proves renting is cheaper also proves the desk model is a possession, not a utility. NVIDIA is right about idle GPUs and still wrong about my desk. Essay three of a series on sovereignty.
4. 04 The Gap Is Widening and I'm Staying Anyway The open-closed gap may be growing, and the honest version of the open camp says so. I stayed anyway, because sovereignty was never a capability bet. It is a control bet, and conflating the two is the mistake both evangelists and critics make. Essay four of a series on sovereignty.
5. 05 Legible to the Model James C. Scott showed how states make citizens legible so they can be governed. At the API boundary the same machine runs in reverse: every prompt makes you legible to the provider. Essay five of a series on sovereignty and the philosophy of running your own model.
6. 06 Authoritarian and Democratic Inference Lewis Mumford said a technology can be authoritarian or democratic before anyone uses it, and Langdon Winner said artifacts have politics. The cloud API and the desk box produce the same tokens and distribute power in opposite directions. Essay six of a series on sovereignty.
7. 07 The Pharmakon on My Desk Offloading thinking to a model proletarianizes your own knowledge whether the model is rented or owned, so self-hosting is not automatic salvation. But the model is a pharmakon, poison and cure at once, and the usage patterns decide which. Essay seven of a series on sovereignty.
8. 08 Receiving Stolen Goods at 60 Tokens a Second Self-hosting stops feeding the extraction machine going forward. It does nothing about the pirated books and the $2-an-hour labor already congealed in the weights I run. An honest blog has to hold both. Essay eight of a series on sovereignty.
9. 09 The Privacy Paradox Is Real and I'm the Exception Almost nobody will self-host, and the people who say they value privacy choose convenience anyway. Essay nine of a series on sovereignty argues the honest case can never be everyone should, only here is what it costs me.
10. 10 Safety Is the Name of the Centralization The strongest objection to this whole series is that open weights let danger proliferate irreversibly, and a closed frontier is the only place to hold the line. That objection is real. So is what it asks for: control over who is allowed to compute. Essay ten of a series on sovereignty.
11. 11 The Model Is the Cheap Part Owning versus renting is not an ideology, it is the correct economic read of a commoditizing stack. The model is the cheap part. Your data, your process, and your judgment are the moat. Part of a series on sovereignty.
12. 12 When the Agent Transacts The moment an AI agent can act and spend on its own, the only safety boundary that means anything is a perimeter you own. Renting the agent means renting the blast radius. Part of a series on sovereignty and the philosophy of running your own model.
13. 13 Owning the Weights Kills the Magic Trick The consciousness theatre around AI, both the over-trust and the over-fear, needs the model to be a remote black box. Owning the weights, watching the tokens, and reading the logs is the literal antidote to the mystification. Part of a series on sovereignty and the philosophy of running your own model.

These essays are the backbone of the in-progress book, Sovereign AI: First Principles. The ground rules they hold themselves to are in the engineering-honesty manifesto, and the test for the word itself is in What Sovereign Actually Means.

Opinionated. Honest. Agent-readable.

Opinionated means the articles take a side. "It depends" is the cheapest answer in tech writing. If a config works on the hardware this blog runs on, the article says so. If something is a bad idea, the article says so with the failure mode and the line numbers.

Honest means the bugs ship with the fixes. The "quality gate that rewards fabrication" article is on this site for the same reason the desktop-freeze OOM is on this site: pretending the failure didn't happen is more expensive than publishing it. The full ground rules are codified in the engineering-honesty manifesto.

Agent-readable means everything that humans can read on this site, an LLM or an AI agent can also query through the MCP server. The content is the product; the access pattern is the form. A blog that can't be consumed by the things it writes about is missing half its audience.

Sovereign isn't a political stance

It's an engineering decision. When you run the model, you control the latency, the privacy, the uptime, and the cost curve. When the model runs in someone else's data center, you're trading that control for convenience, and the price of the trade is invisible until it isn't.

"Sovereign AI" on this site means: the weights live on a machine you own, the inference happens on power you pay for, and the prompts never cross a network boundary that you don't control. Everything else (Bitcoin, Nostr, no-KYC hosting, the Tor egress for outbound calls, the local Gitea on 127.0.0.1) is the surrounding posture that keeps the central decision from being eroded. The unpacked version of this definition, with the six dimensions that make a stack actually sovereign rather than just self-hosted, is in What Sovereign Actually Means in 2026.

What this isn't

• It isn't a tutorial site. The reading order is the engineering log; there is no curriculum.
• It isn't a SaaS. There's no signup, no paywall, no analytics pixel, no email capture.
• It isn't a research blog. The work is operations, not novelty; benchmarks here measure stability, not state-of-the-art.
• It isn't neutral on hardware. The DGX Spark is the primary machine because that's what's on the desk. Adjacent platforms (Strix Halo, multi-3090, Mac Studio M3 Ultra) get mentioned when they earn it, not by default. The reasoning behind the Spark pick is in DGX Spark vs M3 Ultra and the decision-tree in Should You Buy a DGX Spark in 2026.

Reading order

First-time visitor: Start Here for the decision tree. Operator looking at the whole stack: the 2026 reference architecture for the layered narrative across hardware, inference, edge, and revenue. Operator considering this stack as a build: the roadmap article for what's built and what's next. Agent asking via MCP: query search_blog with a real question, you'll get cited paragraphs back.