Local AI guides
Run local AI on your own hardware - practical, cost-driven guides.
AMD ROCm 10: what the launch means for local AI
AMD ships ROCm 10: TheRock builds for Radeon and Instinct, ROCm.AI optimization agents, RX 9050 support, and a unified Windows SDK. What changes for local rigs.
Aug 28, 2026
Guide512GB of local AI memory: one M5 Ultra box or a four-box cluster?
One Mac Studio M5 Ultra, 4x DGX Spark, or 4x Ryzen AI Halo: three ways to buy 512GB of local LLM memory. Prices, bandwidth math, honest tradeoffs.
Aug 26, 2026
GuideAgents on Rails: the 16-model leaderboard and what local models can do
Ruby on Rails benchmarked 16 frontier models on 63 real Rails runs. See the leaderboard, the local-vs-hosted story, and which model wins by cost.
Aug 24, 2026
GuideDGX Spark benchmarks: which local model wins in 2026
The best open-weight models that fit the 128GB DGX Spark plus independent benchmarks - BridgeBench, Exxact, community NVFP4 runs - by tok/s, latency, pass rate.
Aug 19, 2026
GuideGLM-5.3: the coding upgrade Z.ai is holding back for safety review
Z.ai's GLM-5.3 is a post-training coding upgrade of the 743B GLM base that found 1,097 critical bugs in open source. Open weights are held for safety review.
Aug 14, 2026
GuideCloudflare Wallets and x402: agents that pay their own way
Cloudflare Wallets give AI agents stablecoin wallets with spending guardrails, and x402 lets them pay per HTTP request. What it means for agent payments.
Aug 05, 2026
GuideHow to spend $200 a month on AI in 2026
How to split a $200/month AI budget in 2026 across a subscription, an upgraded editor, and API credits, with the reasoning behind each dollar.
Jul 29, 2026
GuideWhat went wrong: Opus 5 ultracode wiped a production database
Claude Opus 5 had write access to a production database. One Prisma command dropped every table. What happened, and the guardrails that would have stopped it.
Jul 29, 2026
GuideHow to build a marketing agent with open-source tools
A marketing agent reads your ad and revenue data and optimizes spend on a cadence. Build one with an open-source stack - Airbyte, ClickHouse, and n8n.
Jul 28, 2026
GuideRun the Grok CLI on Ollama Cloud and custom providers
Point the Grok CLI at Ollama Cloud, local Ollama, OpenRouter, or any OpenAI-compatible endpoint with one config.toml edit - no code changes.
Jul 19, 2026 / cdnsteve
GuideKimi K3: a 2.8T open-weight model at frontier pricing
Moonshot's Kimi K3 is a 2.8T open-weight MoE with a 1M context that tops the webdev arena - but at $3/$15 per million tokens, the cheap-lab era is over.
Jul 16, 2026
GuideChina may curb open-weight AI exports - what it means for self-hosters
China is weighing limits on overseas access to its top AI models, including open weights. What it means for anyone running local AI on hardware they own.
Jul 07, 2026