// blog
Field notes from the local AI frontier
- deployment We Docker-Deployed Every n8n Alternative. The Results Don't Match the Listicles.
Six self-hosted workflow automation tools deployed on identical hardware with docker compose — real deploy friction, idle RAM, and gotchas for n8n, Activepieces, Windmill, Automatisch, Node-RED, and Huginn.
- benchmarks We Tested AI Agent Harnesses on Our Own GPUs. One Claimed Success Without Doing Anything.
dsh and TrueForge, run head-to-head on a self-hosted dual-RTX-3090 rig with a local 27B model and a cloud endpoint — identical 5-task suite, filesystem-verified results, and a cost-per-task column the API-only reviews can't copy.
- deployment Self-Hosting an LLM Without It Phoning Home: A Verified Telemetry Hardening Guide
Self-hosted does not mean silent. A layer-by-layer audit of a real llama.cpp + LiteLLM + AdGuard stack: what phones home by default, the exact opt-outs, and network-level enforcement as a backstop — verified with DNS logs, not docs.
- news Generative Engine Optimization for Docs and Developer Sites
A practitioner's GEO guide for documentation and developer sites — what actually gets you cited by ChatGPT, Perplexity, and AI Overviews, what to skip, and how to measure citations without enterprise tools.
- pipeline Self-Hosted vs API: The Real Cost Breakdown Nobody Shows You
Comparing self-hosted LLM costs to commercial APIs honestly: break-even math, hidden costs on both sides, and why hybrid pipelines win for most teams.
- deployment How Much GPU Do You Actually Need to Self-Host an LLM?
A practical VRAM sizing guide for running large language models locally: model size, quantization, context length, and concurrency — with real numbers instead of guesswork.
- ai-search llms.txt Implementation on Astro + Cloudflare Workers: What We Actually Shipped
A first-hand guide to adding llms.txt to an Astro site on Cloudflare Workers — file structure, content-type gotchas, cache behavior, and how to verify AI crawlers can parse it.