I built Hindsight as a self-improvement skill for AI coding assistants. At the end of a session it reviews the tasks, corrections, and dead ends, then keeps only the lessons that would change how future work gets handled.

I watch nearly every episode, and ModelDeck is a menu bar app that cleanly shows how much of your Claude, ChatGPT, and soon Grok subscriptions you have left, plus a breakdown by project, session, and task so you can see what’s burning through usage and stretch it further.

I finally have this AI memory system and codebase analyzer to a point it’s legitimately helping my sessions be more successful, efficient, and reliable. It has a little integration with graphify for people using that as well.

I follow your GitHub repo breakdowns, especially the episodes on local AI tools and practical agent workflows. Rta-Smriti Brain is now v1.0.2-alpha: local-first project memory that keeps an evidence-backed view of a project’s current reality so a new session doesn’t start from scratch — the name isn’t an acronym, Rta is cosmic order and truth in the Rigveda, Smriti is memory.

This is my first GitHub submission. I’m new to the AI space — I learned GitHub a couple of weeks ago and spent about four weeks immersed in tutorials before building this to solve my own struggle with managing YouTube subscriptions. It uses Andrej Karpathy’s LLM Wiki framework: raw data into a local staging folder, processed into structured Markdown, with a prompt asking whether you want a local search to cut the token overhead.

You asked publicly for GitHub repos and creators worth featuring. This is a client-installable agent skill, not a SaaS dashboard: /reimagine-it reads the source file and redesigns from its own nouns, dates, and colors instead of inventing a vibe.

Radiant is a native Mac app where multiple AI agents share one conversation and consult each other mid-task, backed by a server you can reach from your phone and other Macs. It signs in with subscriptions instead of API keys, runs local models with no key at all, and connects to Hermes and OpenClaw agents already on your machine.

Kiki is a free, fully local voice-intelligence app for macOS that combines system-wide dictation, meeting capture, recording transcription, and voice creation in one private tool. Unlike subscription dictation apps such as Wispr Flow, it requires no account and keeps speech and transcripts on your Mac.

Everyone in the agent space says “autonomous” and nobody defines it. AAS scores a system on a 0–5 scale twice — Active and Ambient — and reports them side by side. Claude Code comes out at 2.29 active and 0.57 ambient; Manus at 2.43 and 0.71; Siri at 1.71 and 0.29. The tools that feel the most agentic in use are the ones that go closest to zero the moment you stop typing. It’s designed to be rated independently — if you score something and land somewhere different than I did, I want to hear it.

I built WhatsKept as a way to query my WhatsApp through AI from anywhere — not only text, but image descriptions, voice-note transcripts, and PDF OCR. It pulls ChatStorage.sql from an iPhone backup, enriches it, and sets up a live MCP server, with install done by prompts so you don’t have to read docs or learn CLI flags.

I built it out of pure necessity. As a stock media creator, the most tedious part was sitting there manually writing titles and keywords for dozens of images and videos, so I built Descify to generate the metadata and embed it directly into the files and skip the tagging grind.