AI that matters, from the architect's desk. Curated and engineered by Saaket Varma, PhD — no hype, just signal.
HOW TO READ THIS Read top to bottom: the GLM 5.2 chip releases its weight file, which forks past the crossed-out API toll gate along the open route and lands on your own server.
Zhipu shipped GLM 5.2 and it rocketed to the top of Hacker News — another open-weight model claiming frontier-class quality from outside the big US labs. The signal that matters isn't the benchmark bragging; it's that high-capability models you can self-host keep arriving faster, which puts real pricing and lock-in pressure on the closed incumbents. If you run inference at any scale, add GLM 5.2 to your eval set this week and measure it against your current closed model on your own tasks — the gap is narrow enough now that the bill, not the leaderboard, may decide it.
HOW TO READ THIS Read top to bottom: Addy Osmani publishes a repo of Markdown skill files, any coding agent loads one to gain a capability, and the repo spreads to 59K+ stars.
A fast-trending library of production-grade engineering skills for coding agents — the clearest sign yet that 'skills' are becoming the reusable unit of agentic dev work.
HOW TO READ THIS Read top to bottom: Crosstalk Solutions ships a boxed AI kit, the model weights live inside it, the cloud link is severed while prompts and answers stay local, and the repo has drawn 30K+ stars.
A self-contained, fully offline survival computer with local AI baked in — a concrete look at capable models running where the cloud can't reach.
HOW TO READ THIS Read top to bottom: one maintainer publishes OpenMed, the open stack of models and tools tags drugs, doses and diseases in clinical text, and any hospital, lab or laptop can run it for free.
An open-source healthcare AI toolkit climbing the charts — regulated, high-stakes domains are starting to build their own open stacks instead of waiting on closed vendors.
A fast KV-cache layer that cuts latency and cost for repeated LLM context — infra you feel directly on your inference bill.
An open AI-driven development agent that writes, runs, and iterates on code end to end.
One control plane to run and scale AI workloads across any cloud or Kubernetes cluster — vendor-neutral compute management.
Andrew Ng's unified interface over multiple GenAI providers, so swapping models is a one-line change.
A lightweight coding agent built for open models like DeepSeek, Kimi, and Qwen — pairs naturally with the self-host shift.
The Sequence A tidy week-in-AI recap: Anthropic ships, Apple borrows, Musk lists, Bezos builds.
SemiAnalysis The case that Intel should raise capital — a clear-eyed read on where the compute supply chain is straining.