2026-11-14 Obsidian Starter Kit v6 is out
Version 6 of the Obsidian Starter Kit is out. It's my own Obsidian vault, packaged so you can start from it: a knowledge system for your whole life. Less in your head, more done by your AI.
Short notes, quick updates, and things I find worth sharing. For the full graph of my thinking, browse my public notes. Subscribe via RSS.
Version 6 of the Obsidian Starter Kit is out. It's my own Obsidian vault, packaged so you can start from it: a knowledge system for your whole life. Less in your head, more done by your AI.
Claude Meetings lets Claude join your video calls as a visible participant. Announced by Anthropic on 2026-11-12, in beta for Team and Enterprise plans (rolling out gradually), and Google Meet only for now (other platforms are "coming soon", with no names or dates).
I've just shipped v2.3.0 of Claude Epic Status Line, my status line for Claude Code. It started with a bug on my Windows machine: the status line showed 354k of context and 244 requests, while the session was already at 359k and 247.
The Gemini Decisions API is Google's decision model) endpoint, announced on 11 November 2026 by Google DeepMind. It runs a new model, Gemini Decision 1 (`gemini-decision-1`). You give it an input and a set of bounded questions, and instead of writing an answer token by token, it scores every option
Gemini 4 Hydrogen is Google DeepMind's third Gemini 4 model, released on 12 November 2026, after Gemini 4 Argon and Gemini 4 Carbon. Argon and Carbon are the big, slow, expensive frontier models. Hydrogen is the fast one: Google calls it "our daily driver for agentic workflows", built for everyday c
Anthropic released Claude Fable 5.2 and Mythos 5.2 yesterday (2026-11-09). I've just shipped v2.2.1 of Claude Epic Status Line, my status line for Claude Code, so that it shows the right per-turn cost for both.
Translating a compiled program's machine code into source code for another platform, ahead of time, then compiling it into a native program. How console games get native PC ports in weeks instead of years.
An open-source library I built to turn small text blocks into music notation: chord diagrams for guitar and piano, keyboards, fretboards, chord charts, tabs and scores, strumming patterns, progressions and the circle of fifths, drawn as SVG and HTML. Every diagram plays, and copies as an image. You
How to architect a Personal Knowledge Management (PKM)) system that AI can consume as context. Not just "use Obsidian." The specific design principles that make a PKM system AI-accessible.
Yesterday (9 November), Anthropic released Claude Fable 5.2 and Claude Mythos 5.2, ten weeks after Claude Fable 5.1. Same structure as before: one model, two levels of safeguards. Fable 5.2 is for everyone, and Mythos 5.2 is for vetted cybersecurity and life sciences organizations.
On 7 November, Shane Melaugh posted this on X: "is Opus 5.5 nerfed? I just had Gemini 4 Argon analyze 1,400 Claude Code sessions and it found that on November 3, something definitely happened. the model name didn't change, but the way it does reasoning took a sharp turn." The post got 582K views, an
The IPCRA method is the organization system of Eliott Meunier, the French creator behind Atomic Thinking. It gives your second brain five top-level folders: Inbox, Projects (Projets), Caps (Casquettes, i.e., "hats"), Resources (Ressources) and Archives. I see it as a tiny ontology for knowledge and
Pele is an Open Source library (MIT) that renders Mermaid diagrams to SVG. It comes from the Obsidian team and was created by Steph Ango (kepano), the CEO of Obsidian. The goal is stated plainly on the website: render diagrams quickly in Obsidian and make them match your theme.
Most PKM methods solve one problem. The PARA method and the IPCRA Method tell you where to file a note. The Zettelkasten method tells you how to turn what you read into your own thinking. Getting Things Done (GTD)) tells you how to get your commitments out of your head. Each one is useful, but a lif
A project context note is a short note, one per project and one per ongoing responsibility, that an AI reads first, before it does anything else in a session. It's the concrete artifact behind Project Context Management): PCM covers everything an agent should know about a project; this note is where
Latent reasoning means letting a LLM) think in vectors instead of words. A normal reasoning model writes its chain of thought as tokens: it picks a word, feeds it back in, picks the next one. A latent reasoner skips the "pick a word" step and feeds its internal state (a long list of numbers, the hid
Nemotron 3.5 Transcribe is NVIDIA's open-weight speech recognition) model, published on 28 October 2026. What sets it apart from classic ASR models like Whisper or Parakeet V3 is that it's promptable: you tell it in plain language how you want the transcript (verbatim or cleaned up, a list of terms
Copilot Fusion is the "model" you pick in GitHub Copilot when you don't want to pick a model. You select Copilot Fusion and an effort level in the model picker, and for each prompt Copilot decides which models to call and how to combine them. You get one answer back.
Cua (https://cua.ai, `trycua` on GitHub) gives AI agents computers they can use: full desktops, with apps, a browser and a shell, that an agent drives and that you can watch. Its pitch, from the Spaces SDK launch on 19 October 2026: the big labs now give their bots their own computers (Claude Duet,
Agent progress summaries are the short status lines (3 to 5 words, like "Running the test suite") that Claude Code shows next to each background agent in the agents panel. To write each line, Claude Code sends the agent's full transcript to the agent's own model, usually Opus, about every 30 seconds
Genie 4 is a world model from Google DeepMind, announced on 27 October 2026 by Raia Hadsell, Diego Rivas and Hang Qi. You describe a world in text, and Genie generates it as video you can move through and change in real time. Think of it as a game engine where the "rendering" is a neural network tha
Kimi K3.1 Pro is the first model of Moonshot AI's K3.1 series, announced around 23 October 2026. It's Kimi K3 with a lot more post-training compute and one architecture change made during post-training (see "The weights" below): same 3T-class mixture of experts (2.8T parameters, 104B active, per the
Gemini 4 Carbon is Google DeepMind's second Gemini 4 model, announced on 28 October 2026, two weeks after Gemini 4 Argon became generally available. Google calls it "our strongest model with state-of-the-art cybersecurity and long horizon capabilities". It's built for long-running engineering and co
Context inertia is when an AI agent hyper-focuses on a hypothesis, a subtask or a goal, and keeps pushing on it long after it should have stopped. The longer it works on something, the more its context is full of that something, and the harder it becomes to see anything else.
Context clash is "when you accrue new information and tools in your context that conflicts with other information in the context" (Drew Breunig, "How Long Contexts Fail", June 2025). Two parts of the context say different things, and the model has to pick, often silently and not always the same way.
Fighting the weights is when "the model doesn't do what you ask because you're working against what it was trained to do" (Drew Breunig, 2026). The instruction is clear and in the context. The model still drifts back to its trained habit.
Strategic forgetting is choosing what an AI agent should forget, and when. It works at two time scales:
On Friday (31 October), Andrej Karpathy opened YouTube, saw "an enshittified wall of slop", and decided to build his own. Two hours later, HackerTube was live: a YouTube clone that only contains videos recommended in Hacker News comments.
The context signal-to-noise ratio (context SNR) is the proportion of task-relevant information versus irrelevant information within an AI agent's context window. It's the core metric that Context Engineering optimizes.
Is Agentic is a free web tool that scores how ready a website is for AI Agents. You submit a public URL and get back a report with a score, the evidence it observed, and the fixes it recommends. Scans are run and scored by "Ora"; the site says nothing else about who makes it.
Agent Experience (AX) is the experience AI Agents have as users of a product or platform: how easily they discover what it does, how reliably they can operate it, and how well they recover when something goes wrong. Mathias Biilmann, CEO and co-founder of Netlify, introduced the term on 28 January 2
Deferred classification means capturing information now and deciding what it is and where it goes later. You put things down without a title or a category, and you classify them in a separate step.
The OpenAI CLI (`openai`) is "The official command-line tool for the OpenAI REST API", published by OpenAI in October 2026. It's a CLI) for calling the API from a terminal or a shell script: generate images, upload files, submit batches, read usage. Apache 2.0, written in Go.
Thinkrail is an open-source agentic IDE from JetBrains, built around Pi (the coding agent by Mario Zechner). It's a JetBrains incubator project, licensed under Apache 2.0. The README pitches it as "the agentic IDE that gets better every time you use it", and the tagline is "Vibe code with pi in a li
Cloudflare Quick Tunnels no longer have to be open to anyone who gets the link. Since `cloudflared` 2026.9.3, you can add `--allowed-mail` to the command, and only the email addresses or domains you list can get in. You still don't need an account or a domain, and it's still free.
GPT-6.1 Terra is OpenAI's model between GPT-6 Luna and GPT-6.1 Sol. It came out in late October 2026, about a month after the GPT-6 family launched without a Terra at all. When GPT-6 Sol and Luna replaced the GPT-5.6 models on 22 September, GPT-5.6 Terra got no successor, and the common reading on H
Microsoft Autopilot is Microsoft's "persistent, proactive and personal agent that keeps working even when you're not". You give it a name, a role and a goal, and it works inside your Microsoft 365 tenant without waiting for prompts: it watches channels, follows up on threads, runs recurring work and
An Obsidian plugin for learning an instrument in your notes: log practice in under a minute, run a practice timer, get a plan for the time you have, and let spaced review tell you what's slipping. Guitar and piano first, any instrument after.
Pi pod runs sessions of the Pi coding agent in isolated sandboxes ("pods") on a server you host, and you compose the environments they run in. Its author, edverma2, shared it on Hacker News as a Show HN in October 2026, where it got 121 points and 46 comments.
I've released two updates of the Kanban Action Planner plugin for Obsidian today: 2.7.1 and 2.8.0. The big one is 2.8.0: time tracking you can start without opening a board, new ways to filter your cards, and new cards that show up the moment you create them. 2.7.1 is a cleanup release; nothing chan
The findability stack is my own model of the different layers of findability in a personal knowledge base. I couldn't find any published model that describes them, so I made one. Each layer is a fallback for the one above it, and resurfacing runs across all of them.
Google Docs can now open, edit and share Markdown files without turning them into Google Docs. You double-click a `.md` file in Google Drive, it opens in Docs, and it stays a `.md` file. Drive also gets a rendered preview for Markdown files.
LangChain Deep Agents is an open-source "batteries-included" agent harness built on top of LangGraph. It provides a pre-configured, opinionated agent system for complex, long-horizon tasks. Ships as a Python SDK (`deepagents`), a TypeScript port (`deepagents.js`) and a terminal coding agent now call
Claude Haiku 5.5 is Anthropic's small, fast Claude model, released on 7 October 2026. It's the third model of the Claude 5.5 family, after Claude Opus 5.5 (22 September) and Claude Sonnet 5.5 (28 September), and it replaces Claude Haiku 4.5, which was almost a year old. Model ID: `claude-haiku-5-5`
OpenShell is a safe, private runtime for autonomous AI Agents built by NVIDIA. It provides sandboxed execution environments using Docker and Kubernetes (K3s) to isolate AI agents while protecting user data, credentials, and infrastructure through declarative policy enforcement.
State-of-the-art open-weight general-purpose multimodal model by Mistral AI, released December 2, 2025. First Mistral mixture-of-experts model since the Mixtral series. Released under the Apache 2.0 License (both base and instruct versions).
Principles and patterns for writing effective, maintainable, and reusable AI Agent Skills. Drawn from practical experience building skill libraries.
Claude Code now compacts long sessions while you're away. Leave a big conversation alone for about an hour and, when you come back, the conversation shows `Compacted while idle, before the prompt cache expired`. From that point on, Claude works from a summary of what you did together.
On Wednesday (21 October), Anthropic announced Claude Duet, a "proactive agent" in research preview. It runs on Claude Opus 5.5, works in its own cloud computer, and keeps going while you're away.
Renaming a note in my vault could empty its lists. No error, no warning. When I went looking, 18 notes had already lost them.
A clear roadmap to building your own knowledge system. Subscribe and get it straight to your inbox.
6,000+ readers. No spam. Unsubscribe anytime.
Every week, I share practical systems, tools, and frameworks to help you organize your knowledge, leverage AI, and create consistently. No fluff. No spam.
6,000+ readers already subscribed ✨
Every Knowii tier includes the private community, live events, expert access, monthly challenges, a 3.8K+ concept knowledge base and my ~20K published notes. Pick the depth that fits you:
€14.99/quarter
€59.99/quarter
€119.99/quarter