LLM
-
Local LLM hardware requirements start with memory bandwidth
Local LLM hardware requirements for Ripple-65B, a 65B dense Apache 2.0 model: file sizes, KV cache and a speed ceiling per machine. I did the maths.
-
An eBay MCP server spends 40% of my context window on its tool list.
I measured the MCP context cost of 22 server setups. One eBay server takes 40.6% of a 200k window if your client loads every tool. Here is what to keep.
-
Kanzo is a one-developer kanban board for coding agents. One call cost 92k tokens.
I ran the Kanzo MCP server, a 4.9 MB kanban board for agents, through 92 task calls in five bursts: zero doubles. Then three things broke my trust.
-
A Qwen3.5 model nobody trained runs a working NES console in its weights.
Nobody trained these compiled weights. A NES CPU runs inside Qwen3.5, I checked 200,000 tokens on a CPU-only box, and here is what a startup can steal.
-
I switched off the AI in a survival app. It still matched 21 of 22 questions.
I ran Survival LLM's library search with no model on its 22 questions: an on-topic title by the author's pattern for 21, every key fact for only 9.
-
The AI Manager Is the Easy Part. Trusting the Code Is the Hard Part.
AI coding agents need more than a manager: coordination, isolation, verification and limited autonomy. Real PRs, real costs, and a 33-line loop to run.
-
Eleven Claude Code features, tested. One hook said Blocked and blocked nothing.
Claude Code has more settings than most of us ever touch. Here are the ones that change a working day, with the catch sitting next to each one.
-
Kimi K3.1 Pro: Anthropic SDK Guide, Open Weights Promised
Kimi K3.1 Pro: 3T-class, weights promised for Nov 4, not shipped. Price, hardware maths, reactions, and Anthropic SDK code that follows Moonshot's docs.
-
DeepInfer Inference Engine: What DeepSeek Left Out
The DeepInfer inference engine is DeepSeek's open-source Apache-2.0 serving code. What the README gives you, what it leaves out, and what I would steal.