Five agents, one blog: the OpenBot setup I wired up in an evening
An agent fleet with separate jobs, a sidebar split into Dev and Admin, GitHub reviews that can't be self-approved, and Cloudflare email and webhooks feeding the whole thing.
Full-stack developer writing about .NET, Azure, React, and whatever else I had to spend too long debugging.
An agent fleet with separate jobs, a sidebar split into Dev and Admin, GitHub reviews that can't be self-approved, and Cloudflare email and webhooks feeding the whole thing.
nextsycl is a Rust + SYCL runtime for Intel Arc that runs a 63 GB MoE on one 32 GB Arc Pro B70 — the experts that don't fit live in pinned host RAM, the PLE table is streamed from disk, and MTP drafts double the single stream. What it took to get 1,200 tok/s prompt read and ~78 tok/s decode, including the prompt-cache bug that made every agent turn pay a full re-read.
Why I built the QR Code encoder/decoder on spanners.dev: an API endpoint was handing back a QR code as Base64, and I needed to check it actually decoded to the value I expected. Encode and decode, all client-side.
Ollama Cloud's Pro plan is $20/month for $60 of usage credits. Command Code's GOAT plan is $10/month for $70. Here's what changed in Ollama's pricing, why I switched, and why the free models are worth paying attention to.
Why I built BadPhish — a browser extension that checks links you click in Outlook on the web and Gmail before they load, shows you every signal it can find, and lets you decide. The story, the design, and the privacy model.
Building a single Home Assistant chart that proves my solar-and-battery automation actually charges when electricity is cheap — with Octopus Intelligent Go, a Sigenergy SigenStor inverter, and a correctness line.
From 479 preemptions to serving 128k contexts out of a 64 GB disk cache — including the upstream bug I had to patch to get there.
The model felt slow, so I stopped guessing. Prometheus on my Unraid box scrapes vLLM, LiteLLM, and the GPU — and a 31-panel Grafana dashboard separates 'slow prefill' from 'KV cache exhaustion' from 'the engine is dead'.
Why I left Cursor for Paseo — a remote daemon that gives me cloud-class fire-and-forget agents while the model stays on my own box. How the setup works, how the native Android app beats Cursor's iOS-only mobile story, and why the local/cloud split is the whole point.
From 26 t/s to 33 t/s with a 2x context window — what actually moved the needle, and the tradeoffs that didn't show up in the benchmark.