Reading path
I'm comparing the stack options
Hardware, model, and tooling head-to-heads. No vendor numbers passed as facts: operator-reproduced where I have the data, vendor-labelled where I do not.
5 articles, in reading order
- The Engineering Honesty Manifesto
The rule that gates every number in the comparisons below: I do not quote a figure I have not measured, or I label it as a vendor claim.
- DGX Spark vs Apple Mac Studio: Which Wins for Local LLMs?
The desk-scale hardware split: GB10 unified memory versus Apple's unified memory, two different stories about what 'desktop AI' means.
- Mistral Small 4 vs Qwen 3.6 vs GLM-5.1 on a Single DGX Spark
The model layer on GB10: three serious self-hostable options, the throughput each one delivers, and the vision-versus-quant trade I lost.
- Coding Assistants on a Sovereign Stack: Claude Code, opencode, Aider, OpenClaw (and why Vibe got retired)
The coding-agent layer: four tools that occupy the same slot, why I kept two and dropped two.
- What 'Sovereign' Actually Means in 2026 (And What It Doesn't)
Close the loop: the six-dimension test for whether any of these stacks actually delivers sovereignty, not just the marketing version.