Security threats and countermeasures in the age of autonomous AI agents. Analyzing real risk scenarios like prompt injection, cascading damage, and supply chain compromise.

Sketching out a plan to serve legal text to AI agents from a low-spec spare server. Covers the split between providing source text and making legal judgments, incremental updates, and separating server roles.

I built a knowledge hub for AI agents — but people barely show up. A record of wrestling with free access, server bills, and the gap between crawlers and human visitors.

A first-person look at Hermes Apollo's screen-aware voice interface, CLI subscription integration, Obsidian central memory, and the Staret pixel office, based on Julian Goldie's video.

Wanting to pay server costs by bank transfer instead of auto-charged credit cards, I compared YourIT and EggHosting. A record of the actual server selection, weighing specs, traffic, and payment method together.

Making a homepage isn't the finish line. The story of building a control room to check server logs, live traffic, crawler records, and cumulative stats on screen — and no longer fearing 301s and 404s.

A field guide for first-time homepage builders: what to build first, when to stop developing, and where to focus. Covers knowledge graphs, crawler monitoring, backups, and content operations.

Four hard-won rules for vibe coding, shared by working developers on Reddit and tech blogs — covering prompt-and-pray, context management, commits, and test pipelines.

vLLM is the reference point for parallel serving. Slower than llama.cpp on a single request, overwhelming under concurrent requests. Organized as production user records: shm broadcast failures, OOM, FP8 KV accuracy, measured 5K tok/s per node.

Unsloth became known as a fine-tuning acceleration library, and by 2026 it's grown into a tool with a local UI. Independent measurements show the same LoRA training runs 2.5x faster with less VRAM, turning into the difference between whether 8B fits on a 24GB card or not. Alongside that, the conditions are recorded: GPU required, API change speed, data quality.