rotecodefraktion
Next-Concept Modeling: From Sketch to Clear Text
NCM does not write text word by word but as a stack of ever finer codes, one whole level per pass. The technique behind it, the authors' numbers, where the limits of this development lie today, and a second paper that lets agents think without words.
Read more ›colibri Tested: A 744B Model on a Mac with 64 GB
colibri runs huge mixture-of-experts models on ordinary hardware by reading the experts from disk. Measured on an M3 Max with 64 GB, from Qwen3.6 up to GLM-5.2 with 744 billion parameters: it runs, but very slowly, and the speed depends mostly on how many experts fit into RAM.
f451 1.3: Import from BookStack and Export Back
f451 1.3 takes over shelves, books, and pages from BookStack as drafts with a review and writes spaces back to BookStack as a book, from the command line or the browser. It also adds Mermaid diagrams, backlinks, and a review overview.
Generating AI Videos Locally: Wan 2.2 on the Mac
AI videos without the cloud and without a subscription: with the open model Wan 2.2 and MLX, they can be generated locally on a Mac with 64 GB. Setup, compute times, the comparison with PyTorch and a trick for readable text in the frame.
OSS Scanner: Anthropic Scans Open Source for Vulnerabilities, With No Human in Between
The Emperor's New Clothes: f451 Templates
Claude Code Mods: Extensions Running Inside Claude Code
Series
All series ›
n8n — Advanced Topics
Deep dives accompanying the "Getting Started with n8n" series: individual topics that go beyond the basics, from LLM memory to advanced workflow patterns.
A Local Coding Agent with apfel
A series on building a coding agent in Swift whose language model runs entirely on your own Mac — from the apfel CLI to integration with Xcode 27.
Getting Started with n8n
A series on n8n as a self-hosted workflow and agent platform — from the first container to a production-ready pipeline with AI classification and SAP integration.
Commentary
All commentary ›Convenient Fear, Postponed IPO: The AI Warnings from OpenAI and Anthropic
Anthropic and OpenAI call for slowing down AI development, and OpenAI president Brockman sees an AGI era. The concrete danger is documented, the doomsday date is not. The proposed measures benefit those issuing the warnings and do not bind China.
Three Days at the Frontier: The Trump Regime Pulls Fable 5 and Mythos 5
Three days after launch, an export-control directive forces Anthropic to disable Fable 5 and Mythos 5 for all customers. What happened, why, and what it means.
f451
Self-hosted wiki with Git as the single source of truth. Pages are Markdown on Forgejo or GitHub, every change goes through draft and review, AI agents contribute via MCP.
BookBridge
Bidirectional sync between Obsidian and BookStack. Images offline, conflict diff instead of remote wins.