Skip to content

// d-central.tech / ai

Sovereign AI for Plebs

Evaluate the Local AI Stack

AI content should separate model choice, hardware choice, privacy boundary, energy use, heat, noise, storage, drivers, inference runtime, and maintenance burden. The right answer depends on the task, not the largest model name.

D-Central connects AI to its infrastructure expertise through power, heat, local compute, repair-minded operations, privacy, and documentation. ASIC miners remain Bitcoin hardware and should not be described as LLM compute devices.

Benchmark Boundaries

AI benchmark claims should preserve model version, quantization, context length, GPU or CPU, driver, runtime, prompt type, and test date. Leaderboards and vendor claims should be treated as references, not guarantees for a local deployment.

Last reviewed April 16, 2026.

Source Basis

AI pages should cite model cards, project repositories, release notes, hardware vendor specifications, driver/runtime documentation, and D-Central infrastructure experience. Benchmark claims should preserve model version, quantization, context length, hardware, driver, runtime, and test date.

ASIC miners do not run LLM workloads. D-Central connects AI to its practical infrastructure domain through power, heat, privacy, local compute, maintenance, and hardware operations.

Reviewer

Reviewed by D-Central editorial staff with a Bitcoin infrastructure, privacy, hardware, and operations lens. Sensitive data, private keys, customer records, and production secrets should not be loaded into experimental AI stacks.

Freshness Policy

Model releases, licenses, GPU prices, driver support, inference runtimes, and leaderboard results change quickly. AI pages should preserve the model and hardware version used, identify stale benchmarks, and separate local privacy guidance from performance claims.

Last reviewed April 16, 2026. D-Central editorial and repair intake, Montreal, Quebec.

Sovereign AI means running AI models locally on hardware you own, instead of renting cloud access. D-Central builds and curates open-source tools so individual Bitcoiners can run their own models — the self-custody move Bitcoin made for money, applied to compute.

Canadian local AI computers → Bitcoiner manifesto

Quick answer

Sovereign AI means running open-source AI models locally on hardware you own instead of renting cloud access — the self-custody move Bitcoin made for money, applied to compute. D-Central builds and curates open tools so individual Bitcoiners can run their own models, settle inference over Lightning, and keep money, compute, firmware, and energy self-custodied.

Two doors

Canadian boxes, or the Bitcoiner stack

Bitcoin mining is one layer. Local inference is another. The computers below are real SKUs with CAD prices — not vapour workstations.

NVIDIA DGX Spark

$9,449 CAD. 128 GB unified, CUDA, DGX OS. Sourced NVIDIA box, first-boot in Quebec.

AMD Strix Halo 128 GB

$7,349 CAD. Ryzen AI Max+ 395. Not the US-only Micro Center Halo SKU.

Custom GPU inference rig

Quoted. Same shop that used to build GPU mining rigs. Discrete GDDR, owner-serviceable.

Which local AI computer should a Canadian buy?

What this section is

D-Central ships AI content and three live local-inference computers for Canadian buyers, plus the Bitcoiner guides for people who already think like miners. The same open-source instinct behind our mining firmware, DCENT_OS (GPL-3.0 source; guarded experimental artifacts on two exact XIL lanes), is how we approach compute — own it, run it on hardware you control, trust nothing you can’t read. None of this replaces the Bitcoin core of the shop — the AI vertical is additive. Same hashcenter mindset, new compute workload.

TL;DR — the sovereign compute loop

What “sovereign compute” actually means

The sovereign compute loop is a closed circuit a Bitcoiner can own end to end: you hold your own money (Bitcoin), run your own compute (local AI models on hardware you control), on your own firmware (open-source, auditable — led by DCENT_OS), powered by your own energy (the heat your miners already produce), and settle the value directly — machine-to-machine micropayments over Lightning and the L402 protocol, no rented cloud, no middleman, no permission. Money, compute, firmware, energy, settlement — each layer self-custodied, each one more layer decentralized.

Pick your pillar

Five entry points

Every post on this site slots into one of these five. Pick the one that matches where your head is at.

Sovereignty

Why self-hosted AI is the next cypherpunk battle.

Self-Hosting

The pleb's whole-stack guide to running AI locally.

Hardware

DGX Spark, Strix Halo 128 GB, or a custom GPU tower — live Canadian SKUs.

Hashcenter

Hashcenter, not datacenter. Heat, sats, and sovereign intelligence under one roof.

Bitcoin × AI

When public miners left Bitcoin Hashcenters. Why plebs didn't.
New pillar

AI Agents — let the model drive the miner

Where the sovereign compute loop gets hands. Agents like Claude Code and Codex don’t just answer prompts — they control hardware, settle in sats, and (if you’re careless) walk off with your keys. Read these before you wire an LLM to anything that holds money.

Can you run AI on a Bitcoin miner?

The honest answer: an ASIC can’t run AI — but the miner around it can host the agent. Where the line really is.

Control a miner with Claude Code (DCENT_axe MCP)

Wire an agent to your rig over the DCENT_axe MCP server and let it tune, monitor, and report — in plain English.

Miner + Lightning node: sell inference for sats

Close the loop. Meter your local model behind L402 and let agents pay per request over Lightning — no account, no cloud.

The CLAUDE.md wallet-stealing attack surface

Agents read instruction files. A poisoned one can drain a wallet. The threat model every self-hoster needs first.
First time here?

Start here — the five-step path

Read these in order. By the end you’ll have a local model running on your box, a real UI in front of it, and enough theory to pick the right quant.

  1. 1.

    Read the Manifesto

    The narrative anchor. Why sovereign AI is the same move Bitcoin made for money.

  2. 2.

    Read the Pleb's Guide

    Whole-stack overview — models, runners, hardware, the works.

  3. 3.

    Install Ollama

    Ten minutes from zero to a local model answering prompts on your own box.

  4. 4.

    Give it a UI (Open WebUI)

    A ChatGPT-shaped front end that talks to your Ollama node. No cloud.

  5. 5.

    Understand quantization

    GGUF, Q4, Q8, FP16 — what actually fits in your VRAM and why.

  6. 6.

    Dictate locally with DCENT_Voice

    Open-source local dictation: CPU Parakeet by default, Whisper fallback, and no cloud unless you explicitly opt in.

Latest drops

Fresh from the vertical

Newest posts across all six AI categories.

Model library

Open-weight models, catalogued

Long-form model pages — architecture, quantizations, hardware that runs them. Built on the dc_ai_model CPT.

DeepSeek V4

DeepSeek

DeepSeek's April 2026 open-weight pair — V4-Pro at 1.6T total/49B active and V4-Flash at 284B/13B active. MIT license,…

View model →

OpenAI gpt-oss

OpenAI

OpenAI's first open weights since GPT-2 — gpt-oss-120b runs in 80 GB, gpt-oss-20b in 16 GB. Apache 2.0,…

View model →

Qwen 3

Alibaba

Alibaba's May 2025 release — first open family with hybrid reasoning (toggle-able chain of thought), Apache 2.0 across…

View model →

Llama 4 (Scout/Maverick)

Meta

Meta's April 2025 MoE-and-multimodal release, headlined by Scout's 10M-token context window. The pre-announced Behemoth frontier model never shipped…

View model →

Gemma 3

Google

Google DeepMind's March 2025 Gemma family — vision-capable (4B+), 128K context, with official quantization-aware 4-bit variants.

View model →

Mistral Small 3

Mistral AI

Mistral AI's January 2025 24B model — Apache 2.0, competitive with Llama 3.3 70B, fits on a single…

View model →

Full library →

Tools & datasets

Local-AI tools & datasets

The AI vertical’s own free calculators and open datasets — size a model to your GPU, price the electricity, pick a quant, or pull the raw data under CC BY 4.0. Built and maintained by D-Central.

LLM fine-tuning VRAM planner

How much GPU memory to TRAIN a model: full vs LoRA vs QLoRA, exact weights+optimizer math + activation estimate, fit-checked against real GPUs.

Self-hosted LLM & AI-agent security

The OWASP LLM Top 10 (2025) mapped to concrete defenses for self-hosted Ollama/vLLM/RAG/agent setups: prompt injection, excessive agency, supply chain and more.

GPU ↔ Local-LLM fit & sizing

Which local LLM your GPU can actually run, at Q4/Q8/FP16 — now KV-cache-aware for real context.

Local LLM model database

Open-weight models with params, context, license, VRAM and Ollama tag. CSV/JSON, CC BY 4.0.

AI-GPU hardware database

Consumer & datacenter GPUs plus Apple Silicon, ranked by VRAM for inference. CSV/JSON.

Self-hosted AI inference cost calculator

Tokens per second and electricity cost per model, on your own card and power rate.

GGUF quantization quality reference

Bits-per-weight and quality tier for every GGUF quant type — pick Q4 vs Q8 with eyes open.

AI inference accelerators beyond GPUs

NPUs, unified-memory SoCs and open accelerators that can (and cannot) run local LLMs.

Decentralized compute & DePIN networks

Rent GPU compute from decentralized marketplaces — the non-cloud option.

Local AI runtime comparison

Ollama vs llama.cpp vs LM Studio vs vLLM — pick the runner that fits your box.

LLM quantization format comparison

GGUF vs MLX vs EXL3 vs GPTQ vs AWQ vs FP8 — which quantized file your hardware actually wants (ExLlamaV2/AutoGPTQ/AutoAWQ are archived).

Local embedding models for RAG

20 open-source embedding models sorted by VRAM, with dimensions, context, MTEB scores and license.

Self-hosted vector databases

14 self-hostable vector DBs for RAG on index type, hybrid search, quantization and which are genuinely open source.

Local voice-AI models (STT/TTS)

21 speech-to-text & text-to-speech models graded by license, languages (EN/FR), speed and Raspberry-Pi viability.

LLM agent-capability (tool-calling)

Which self-hosted LLMs reliably drive agents and tools — BFCL / tau-bench scored, native tool-format vs grammar-constrained.

Local AI telemetry & air-gap audit

Does your self-hosted AI phone home? Source-verified default endpoints, opt-out env vars and air-gap recipes.

AI/LLM benchmark hub

A capability-organized map of AI-model benchmarks — coding, agentic, reasoning, vision, plus first-party hardware/PCB.
The thesis

Bitcoin mining → AI compute: the bridge

A Bitcoin Hashcenter and an AI hashcenter are the same building told twice: power in, heat out, dense silicon, an operator who already speaks watts and thermal. The hard part of AI compute — the power and the heat — is the part Bitcoin miners solved years ago. This is the path from the room you already run to the inference you can own.

Hardware library — upcoming

GPUs, rigs, and the hashcenter retrofit

A dedicated hardware CPT plus benchmarks database ships in v1. Until then, the foundational reads:

Heads up: a full hardware CPT plus benchmarks taxonomy lands in a later task on the roadmap.

Shoulders of giants

None of this is ours. The entire sovereign-AI stack the plebs now run at home was built by an open ecosystem of researchers, engineers, and labs who released weights, code, and tools under terms anyone can use. D-Central stands on their shoulders and contributes where it can. Named with gratitude:

llama.cpp (Georgi Gerganov) Ollama LM Studio / Element Labs Open WebUI (Timothy J. Baek et al.) Meta (Llama) Google (Gemma) Alibaba (Qwen) Mistral AI DeepSeek Black Forest Labs (FLUX) Stability AI Microsoft (Phi) Hugging Face
[open source] Everything D-Central builds is open-source — led by DCENT_OS, GPL-3.0. Own your money, own your compute. One more layer decentralized.
Source the silicon

Build your hashcenter

Everything we build is open-source (GPL-3.0), led by our mining firmware DCENT_OS. The same shop that powers the Bitcoin side stocks the open-source miners and heat-reuse hardware to start your hashcenter today.

Shop sovereign AI hardware →   Open-source miners   All hardware

Go deeper: Local LLMs in Canada · Distributed compute · Part of the D-Central Sovereignty stack — see all layers →

Canadian local AI computers

Priced SKUs, not vapour workstations. Spark and Strix Halo include first-boot setup in Laval. The GPU tower is quoted from street.

NVIDIA DGX Spark — $9,449 CAD · AMD Strix Halo 128 GB — $7,349 CAD · Custom GPU inference rig (quote)

Bitcoiner mining hardware

This rail is the mining layer: refurbished ASICs and open-source miners from the Laval workshop. An S19 is not a GPU.

Bitmain Antminer S9Bitmain Antminer S9 Price range: $35.00 through $155.00 CADIn Stock — ShopBitmain Antminer S19 ProBitmain Antminer S19 $600.00 CADIn Stock — ShopAntminer Slim EditionAntminer Slim Edition Price range: $560.00 through $745.00 CADIn Stock — ShopAntminer Loki EditionAntminer Loki Edition Price range: $585.00 through $803.00 CADIn Stock — ShopBitaxe open-source ASIC solo Bitcoin miner front viewThe Bitaxe Price range: $184.99 through $224.99 CADIn Stock — Shopminibit-bitaxe-solo-bitcoin-minerThe Modern Minibit Gamma $229.99 CADIn Stock — ShopThe NerdaxeThe Nerdaxe Price range: $169.99 through $199.99 CADIn Stock — Shop

Browse all in-stock miners

Editorial review and limitations

Reviewed by D-Central's mining hardware and ASIC repair editorial team for practical accuracy, buyer risk, repair context, and operational assumptions. Verify current hardware price, stock, network difficulty, BTC price, power rate, shipping, tax, firmware, and device condition before buying, hosting, repairing, or retiring mining hardware.

Last reviewed April 16, 2026. D-Central, Montreal, Quebec.

Related products, repair, and setup paths

Last reviewed April 16, 2026.