Short notes, quick updates, and things I find worth sharing. For the full graph of my thinking, browse my public notes. Subscribe via RSS.
Matt Van Horn — co-founder of the June oven and an early member of the team that became Lyft — has been building in open source lately, and one of his projects is worth your attention: last30days, an AI agent skill that researches any topic across the places people actually talk.
Open Code Review is an open-source, AI-powered CLI for automated code review from Alibaba. It was incubated from Alibaba's internal tooling after two years and tens of thousands of developers, then released to the community. Its distinguishing bet is a hybrid deterministic + agent architecture that
Personal Knowledge Management promises clarity, creativity, and compounding returns on intellectual effort. But the practice is riddled with failure modes that can turn a knowledge system into a productivity sink. These anti-patterns are among the most discussed topics in PKM communities, precisely
If you've ever loved jq for slicing JSON and wished you had the same thing for Markdown, mq) is exactly that.
Quick one: I just shipped a new public wiki dedicated to Personal Knowledge Management (PKM)) — the PKM AI Wiki, live at https://pkm-wiki.knowii.net.
An open-source static-site template I built to turn any cross-linked markdown corpus (typically an LLM Wiki inside an Obsidian vault) into a deployable public site.
While poking around Open Design today, I bumped into one of its plugins worth highlighting on its own: HTML Anything.
I came across Open Design today, and it's worth sharing right away. It's an open-source, AI-native design platform that treats UI generation as an agent workflow instead of a manual canvas exercise. That framing alone got my attention.
Claude Code v2.1.139 ships agent view, opened with `claude agents`. It's a single inbox-style screen for every background Claude Code session you've dispatched, across every project. Rows show what's running, what needs your input, what's done. You dispatch new sessions from the same screen, peek at
In this article, I want to walk you through what Anthropic just shipped on 2026-05-27, what people are already saying about it, and what I think it really means for those of us serious about Personal Knowledge Management (PKM)).
Herdr is a terminal-native agent runtime; "one terminal for the whole herd". It sits between traditional terminal multiplexers (tmux, screen) and desktop agent applications, designed for engineers running multiple AI agents in parallel.
Codex Dynamic Workflows is a Codex CLI skill by DannyMac180 that turns large, ambiguous tasks into supervised AI-agent work. Instead of executing directly, the skill forces planning, delegation, and verification before anything risky happens.
Open-source, model-agnostic platform for building and running autonomous coding agents. Operated by All Hands AI; CEO Robert Brennan, Chief Scientist Graham Neubig (also CMU LTI). Cloud is hosted at `app.all-hands.dev`; core repo is MIT-licensed and community-maintained under the `OpenHands` GitHub
Portent is an open specification for work and personal knowledge bases. It defines a small, opinionated vocabulary of note types, relationships, and lifecycle states so that knowledge bases become "easy for humans and agents to understand." Released under the MIT license by Luca Rossi (founder of Re
If you sync your Obsidian vault through iCloud Drive, startup can be painfully slow because iCloud lazily downloads file contents only when something tries to read them. Obsidian scanning thousands of notes on launch triggers that download cascade.
Out of the box, Obsidian does not include AI features. It fully relies on external AI applications and AI community plugins to provide AI support.
A tool for detecting and removing watermarks added by AI image generators (like DALL-E, Midjourney, Stable Diffusion) to images.
Claudian embeds AI coding agents (Claude Code, Codex CLI, Opencode, and others) directly into Obsidian. The vault becomes the agent's working directory — file read/write, search, bash, and multi-step workflows all work out of the box.
Six months ago Cursor was an AI code editor. Today it is closer to an agent control room that happens to contain an editor. That shift did not arrive in one release. It came across four of them, and the direction is consistent enough to be worth writing down.
Google I/O 2026 happened yesterday, and Google shipped a lot. New models, a reworked developer platform, agents everywhere. Most of a keynote like this is noise unless you actually build with the stuff. So here's the short version: what changed, what matters for knowledge work, and the one date you
For a long time my site effectively had a single RSS feed: the one at /rss/. It worked, but it forced an all-or-nothing choice: subscribe to everything I publish, or nothing.
A Ghost site looks like it has a single RSS feed (the one at `/rss/`), but it actually exposes many, and you can add your own filtered feeds on top. This note captures the process so I can explain it to others.
Artificial Analysis is an independent benchmarking and comparison platform for AI models and inference providers. It tracks frontier models across providers like OpenAI, Anthropic, Google (Gemini), Meta, DeepSeek, Mistral, and xAI; scoring each on quality, speed (tokens/sec), latency (time-to-first-
Google Antigravity is Google's agentic development platform, given its first major platform release at Google I/O 2026 (19 May 2026). It is built around a single agent harness exposed through four surfaces:
A school teacher needed hero images for a school fair: countries, cities, landmarks, the kind of thing you actually want to print and pin on a wall. I spent an afternoon turning the prompts I converged on into two reusable AI skills and pushed them to GitHub so anyone can grab them.
A plugin that publishes notes from an Obsidian vault to a Ghost blog, with reusable presets for tags, newsletters, and frontmatter conventions.
Fire Pass is a $49/month early-access subscription from Fireworks AI that gives unlimited access to Kimi K2.6 Turbo through a dedicated API key. No per-token billing on that model; full 256k context window; standard per-token pricing still applies to every other Fireworks model.
Since shipping v4.0.0 two days ago, I've pushed three patch releases in quick succession: v4.0.1, v4.0.2, and v4.0.3. Here's the short version of what each one brought to the table.
Oh My OpenAgent (omo) is an agent-orchestration harness built on top of OpenCode; it coordinates multiple specialized agents to handle complex development work end-to-end with limited supervision. It was previously called oh-my-opencode.
Plannotator is an open-source review surface for AI coding agents; it lets you annotate the agent's plan before execution and review the resulting diff in a PR-style interface, then send structured feedback back to the agent with one click.
OpenChamber is an open-source desktop and web interface built on top of OpenCode; it packages the agent harness into a multi-surface workspace for "engineering with agents".
Agent view, opened with `claude agents`, is one screen for dispatching and managing many background Claude Code sessions. Each row shows what a session is doing, what it needs, or what it produced — without scrolling through transcripts. Requires v2.1.139+; in research preview, so the UI and shortcu
RTK (Rust Token Killer) is an open source CLI proxy written in Rust that intercepts shell command output and compresses it before it reaches an AI coding agent's Context Window. It sits as a transparent hook between the agent and the underlying tools (git, cargo, npm, docker, kubectl, AWS CLI, etc.)
Andrej Karpathy's term for software in which the runtime is a Large Language Model) and the program is a prompt (in natural language, structured prose, or markdown). The model interprets the prompt against the current input and emits behavior, optionally calling tools to interact with the world.
The Agent Development Lifecycle (ADLC) is the engineering discipline for building, evolving, and operating AI Agents in production. It is the agent-era counterpart to the Software Development Lifecycle (SDLC); the stages share names, but the failure modes, evaluation methods, and operational concern
Also called function calling. The pattern where a LLM) is given a list of available functions/tools (with names, descriptions, and parameter schemas), and the model decides when to invoke them and what arguments to pass. A host (an AI Agent Harness, an SDK, or the model provider's own platform) exec
Software architecture principles transfer to AI skill and agent design with minimal translation. Skills behave like pure functions. Agents behave like microservices. The system forms a dependency graph governed by the same principles that make software maintainable.
The Mistral SDK is the official client library for Mistral AI's API platform, La Plateforme. Sits at the model SDK layer of AI SDKs, peer to Anthropic SDK and OpenAI SDK. Officially supported in Python (`mistralai`) and TypeScript / JavaScript (`@mistralai/mistralai`), with a Go and TypeScript imple
The Obsidian team just published the results of two independent security audits of Obsidian Sync; one by Cure53 (Berlin, October 2024) and one by Trail of Bits (New York, December 2025). Both cover the Sync API, server, and cryptography. All findings have been addressed via remediations validated by
Hermes Agent is Nous Research's open-source, model-agnostic agent harness; distinct from the Hermes LLM line that shares the name. Where Hermes (the model) is a fine-tuned base, Hermes Agent is the surrounding scaffolding that turns any LLM into a long-running personal assistant with skills, memory,
Grok is xAI's Large Language Model (LLM)). Grok is multi-modal. It can generate text, images, etc. It's deeply integrated in the X platform. Users can ask Grok to analyze tweets/threads and to generate content by using `@grok ...`.
New video is up: Levels of AI Context Management — From Generic Answers to Precision Results.
Just shipped Obsidian Starter Kit v4. This is the AI-native release; months in the making.
Official note synchronization service of Obsidian. It's a paid service. Currently (end 2024) costs $4 per month.
Multica is an open-source platform that treats AI coding agents as autonomous team members. You assign tasks to agents the same way you would to human colleagues; they pick the work up, execute it, stream progress, and manage their own lifecycle. The pitch is "project management for human + agent te
The Obsidian team just launched Obsidian Community at community.obsidian.md; a new directory and developer platform for plugins and themes. I've been waiting for this for years. It reshapes plugin discovery and developer compliance, and it frees up serious time for the Obsidian team itself.
Microsoft's new 3D generative AI framework Trellis 2 makes creating high-quality 3D models fast and accessible. The best part? The generated models can be directly imported into Blender and other 3D tools for further refinement, animation, and rendering.
The Obsidian Community is the official platform for discovering, evaluating, and tracking Obsidian plugins and themes. Announced on May 13, 2026, it replaced the previous manual submission process with automated scanning, developer dashboards, and a searchable directory.
HTML is opening up a richer way to explore AI-generated ideas. Not as a replacement for markdown — markdown remains excellent for text-first content — but as another canvas, one where ideas can include animations, interactivity, charts, layout, and structure that text alone can't carry.
HTML is another way to represent and explore AI-generated ideas — alongside markdown, not as a replacement for it. Markdown stays excellent for text-first content: notes, drafts, transient conversation. But HTML opens up a much wider expressive surface for moments when ideas benefit from animation,
A clear roadmap to building your own knowledge system. Subscribe and get it straight to your inbox.
6,000+ readers. No spam. Unsubscribe anytime.
