A W3C proposal enabling web developers to access browser-provided or OS-provided language models directly via a JavaScript API. Supports prompt-completion workflows with stateful sessions, streaming, structured outputs, and multimodal inputs — all running on-device with no cloud dependency.
Just published a new piece — Heavy AI Agents Are an Anti-Pattern: Why Fewer Agents With More Skills Wins.
#pillar-creator-systems
Heavy specialized AI agents are an anti-pattern. Fewer agents with deep skill libraries beat them on every axis that matters: cognitive load, flexibility, context budget, determinism, and maintenance
Tolaria is a free, open-source desktop app for managing markdown knowledge bases, created by Luca Rossi (author of the Refactoring newsletter). Positioned as "a second brain for the AI era", it stores everything as plain Markdown files with YAML) frontmatter on disk, ships a block-based editor with
Qwen3.6-27B is a dense, natively multimodal 27B-parameter open-weight LLM) from the Qwen family (Alibaba Cloud), released on 2026-04-22. It targets the "flagship-quality model that fits on a single high-end consumer GPU" slot in the lineup, sitting alongside the smaller MoE Qwen3.6-35B-A3B and the A
OpenAI Privacy Filter is an open-weight, Apache 2.0-licensed token-classification model published by OpenAI on HuggingFace for detecting and masking personally identifiable information (PII) in text. It is designed for high-throughput data sanitization workflows: feeding LLMs, redacting datasets, on
Fourth-generation flagship release from Deepseek (April 24, 2026). Two open-weight variants — V4-Pro (1.6T total / 49B active parameters) and V4-Flash (284B total / 13B active) — both built on a MoE) architecture, ship with a 1M-token Context Window by default, and fold what was the separate R reaso
DESIGN.md is an open Apache-2.0 format specification from Google Labs for describing a design system to coding agents in a single, version-controllable file. It pairs machine-readable design tokens in YAML front matter with human-readable design rationale in markdown prose — so an LLM) can apply vis
A plugin that turns a structured Obsidian note tree into an actual book; exported to EPUB and PDF.
A family of attention mechanisms that compute interactions between only a subset of token pairs instead of every pair, breaking the O(n²) cost of full self-attention. The dense attention used in vanilla Transformers forces every token to attend to every other token; sparse attention restricts that i
AI models whose trained weights are publicly released, allowing anyone to download, run, fine-tune, and deploy them. "Open weight" is more precise than "open source" because most releases don't include training code or data.
Granite 4.1 is IBM's open-weight LLM) family released in April 2026 under the Apache 2.0 License. It is a deliberate retreat from the MoE) direction taken by Granite 4.0, returning to a decoder-only dense transformer design with no expert routing, no sparse layers, and no extended reasoning chains.
A clear roadmap to building your own knowledge system. Subscribe and get it straight to your inbox.
6,000+ readers. No spam. Unsubscribe anytime.
