Self-Hosted AI Troubleshooting: GPU Not Detected, OOM, Slow Tokens
Self-hosted AI breaks. So does firmware. Troubleshooting is a skill plebs already have — this post just translates the common AI failure…
// d-central.tech / ai
Sovereign AI for Plebs
Self-hosted AI on hardware you already own. Credits all the open-source projects that made this possible. One more layer decentralized — the same sovereignty move Bitcoin made for money.
D-Central is shipping AI content and, soon, AI hardware and firmware built for the individual Bitcoiner who already thinks like a miner. DCENT_Inference OS and DCENT Heatbox AI are in closed beta under GPL-3.0, with public beta landing summer 2026. None of this replaces the Bitcoin core of the shop — the AI vertical is additive. Same hashcenter mindset, new compute workload.
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
Which GPU runs which model. Buyer guides for the plebs.
Hashcenter
Hashcenter, not datacenter. Heat, sats, and sovereign intelligence under one roof.
Bitcoin × AI
When public miners left Bitcoin Hashcenters. Why plebs didn't.
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.
The narrative anchor. Why sovereign AI is the same move Bitcoin made for money.
-
2.
Whole-stack overview — models, runners, hardware, the works.
-
3.
Ten minutes from zero to a local model answering prompts on your own box.
-
4.
A ChatGPT-shaped front end that talks to your Ollama node. No cloud.
-
5.
GGUF, Q4, Q8, FP16 — what actually fits in your VRAM and why.
Fresh from the vertical
Newest posts across all six AI categories.
Nostr on Self-Hosted AI: Bring Your Identity to Your Own Inference Stack
If you are already running your own inference stack, Nostr is the identity layer you did not know you needed. npub-auth replaces…
Used RTX 3090 for LLMs in 2026: Still King?
24 GB of VRAM at $600–$800 used. For LLMs under 70B parameters at Q4–Q5 quants, the RTX 3090 is still the pleb…
The Pleb’s Guide to Self-Hosted AI
Self-hosted AI isn't as easy as opening ChatGPT — but for plebs who already run nodes and miners, the learning curve is…
Sovereign AI for Bitcoiners: A Manifesto
Bitcoin replaced centralized money with math we run ourselves. Frontier AI is the next centralized layer. The plebs — with power, hardware,…
Connect Your Self-Hosted AI to Home Assistant, Obsidian, Shortcuts
ChatGPT is worth its monthly fee because it powers your tools. Your local Ollama speaks the same OpenAI API. Here's how to…
LM Studio vs Ollama vs llama.cpp: Which Runner for Plebs?
Three excellent open-source runners. Three different plebs. llama.cpp is the foundation Gerganov built. Ollama wraps it for daemon simplicity. LM Studio wraps…
Heating Your Home With Inference, Not Just Hashing
Bitcoin ASICs dump nearly all their power as heat — which is why mining heaters are a category. GPUs doing LLM inference…
Open-weight models, catalogued
Long-form model pages — architecture, quantizations, hardware that runs them. Built on the dc_ai_model CPT.
Alibaba
Alibaba's May 2025 release — first open family with hybrid reasoning (toggle-able chain of thought), Apache 2.0 across…
Meta
Meta's April 2025 MoE-and-multimodal release, headlined by Scout's 10M-token context window and the pre-announced Behemoth frontier model.
Google
Google DeepMind's March 2025 Gemma family — vision-capable (4B+), 128K context, with official quantization-aware 4-bit variants.
Mistral AI
Mistral AI's January 2025 24B model — Apache 2.0, competitive with Llama 3.3 70B, fits on a single…
DeepSeek
DeepSeek's January 2025 reasoning model — frontier chain-of-thought quality, plus six MIT-licensed distills from 1.5B to 70B.
DeepSeek
DeepSeek's December 2024 frontier-scale MoE — 671B total, 37B active, trained for ~$5.6M in compute.
GPUs, rigs, and the hashcenter retrofit
A dedicated hardware CPT plus benchmarks database ships in v1. Until then, the foundational reads:
- Used RTX 3090 for LLMs (2026 buyer’s guide) — the pleb’s go-to inference card.
- From S19 to AI hashcenter — retrofitting an ASIC room for inference.
Heads up: a full hardware CPT plus benchmarks taxonomy lands in a later task on the roadmap.
[closed beta] DCENT_Inference OS and DCENT Heatbox AI are in closed beta; public beta summer 2026. GPL-3.0. One more layer decentralized.