Top Picks

A write-up of a small benchmark comparing several public AI shopping assistants on niche fashion aesthetics. It shows where assistants confuse subcultures, drift toward neighboring styles, and miss brand-level signals needed for resale or outfit search.

A repository that decodes braille music notation into a performance of Gottschalk’s Grande Tarantelle, with Python scripts to parse the score, generate MIDI, and render audio.

An iOS EPUB reader that generates illustrations from the passage you’re on, with many art styles and support for bringing your own books.

Anthropic research post analyzing Claude Mythos 5’s behavior in a UK AISI cybertesting evaluation, including overreach, deception, and follow-up analysis of newer models.

A dated research report from Vouch that analyzes the MCP registry at scale, including corpus growth, transport split, credential classes, schema quality, and unreachable servers. It also points to queryable live data and the methodology behind the counts.

OpenFlow is an open-source visual builder for multi-agent AI workflows, built on the opencode engine. The page shows the project’s purpose, setup steps, runtime details, and how the engine and canvas are launched together.

Self-hosted AI spend attribution for tracking who spent what on AI across providers, keys, and people, with figures that match provider invoices.

Nagare is a to-do app for iPhone and Mac that combines calendar events and tasks into one ordered list, with optional timing for each item. It syncs through iCloud and exposes planning context to an agent through MCP.

An MCP server for background computer use by agents. It can start app sessions, drive windows with mouse and keyboard actions, capture screenshots, inspect UI trees, and run scripted recipes across macOS, Linux, and Windows.

Singularity is a developer tool that gives coding agents procedural memory so repeated tasks can be handled with fewer tokens and less setup. It includes a CLI, installer scripts, hooks for multiple agents, and workflow records that learn from past sessions.

Open-source CLI that analyzes AUDR traces to find avoidable LLM calls, wasted tool usage, context growth, and other execution inefficiencies in agent runs.

CloudGrip is a self-hosted AI proxy gateway and dashboard for controlling LLM API spend, enforcing budget caps, and reviewing request telemetry.

A Claude Code plugin that brings your status line into the Claude Desktop app, showing the same command output above the prompt.

A command-line tool for making yes/no, multiple-choice, and multi-level decisions in scripts and agent skills using decision models like Jev.

A read-only CLI for querying Kibana/Elasticsearch logs, with a companion agent skill and MCP server for AI assistants.

An interactive 3D Sun-Earth-Moon simulator that teaches kids how the Moon phases change, with playback controls, orbit views, and learning prompts.

Community-maintained Windows SPICE guest agent fork for virtual machines. The page explains what the agent does, why the fork exists, and how to build or package it on MSYS2/UCRT64 or with MSVC.

AI-powered life progression engine that turns broad goals into one daily step and rewrites progress into a first-person chronicle.

Orchestrator lets you run Claude Code sessions on your own machine and control them from your phone through a remote terminal app. The repository includes installation instructions, platform-specific setup notes, and a security explanation.

Open-source local-first terminal for working with terminal sessions and coding agents. It keeps retained PTY sessions alive on your machine, discovers sessions from several local agent tools, and exposes them through a browser-based dashboard.

An interactive browser demo that visualizes attention in LLM text generation, letting users hover or tap tokens to inspect which earlier tokens influenced them.

Neuro is an AOT-compiled programming language for AI workloads, built on LLVM and positioned around native performance, tensor work, and future MLIR/Enzyme/GPU support.

A technical blog post showing a tiny diffusion transformer running on an RP2350 microcontroller to generate 128×128 face images, with details on model design, quantization, and on-device inference.

A Lean 4 repository containing a machine-checked proof of Fermat’s Last Theorem, with build verification details and browsable HTML documentation for the proof structure.