Top 10 Repos Explained: Archify, Omarchy, OpenMAIC, and More
Two hosts screen-share their way through this week's ten most useful GitHub repos, then react live on air to seven reader-submitted projects.
Posted
3 days ago
Duration
Format
Listicle
educational
Views
54.1K
794 likes
57 · 43
Big Idea
The argument in one line.
The most useful new GitHub repos aren't building new categories of software, they're making existing expensive or proprietary categories (SEO tools, learning platforms, operating systems, forecasting models) open, customizable, and runnable on hardware or a budget you already have.
Who This Is For
Read if. Skip if.
READ IF YOU ARE…
You track AI and dev-tool GitHub repos weekly and want a filtered, explained shortlist instead of scrolling trending yourself.
You're deciding whether to self-host an SEO stack, a small language model, or a Linux setup instead of paying for a SaaS equivalent.
You build in public and want to see how two hosts evaluate a first-time contributor's repo on air, warts and all.
You use Claude Code, Codex, or another coding agent and want plugins that give it memory, current package knowledge, or a review step.
SKIP IF…
You want deep technical setup instructions for any single tool. Each repo gets 2-4 minutes of reaction, not a walkthrough.
You're looking for enterprise SaaS coverage. Every tool here is open-source, self-hostable, or free to run.
TL;DR
The full version, fast.
This episode of The Next New Thing runs through ten trending GitHub repos: an AI system-diagram generator (Archify), a browser-based satellite/plane/ship tracker (God's Eye View), an AI classroom builder (OpenMAIC), Zapier's new SDK, a Go-freshness plugin for coding agents, an unauthorized OpenAI/Gemini fork of Claude Code (OpenClaude), 160+ scientific research skills, a from-scratch small-language-model trainer (MiniMind), a $10/month open-source SEO platform (OpenSEO), DHH's AI-native Linux distro (Omarchy), and Google's TimesFM forecasting model. The second half reacts live to seven audience-submitted repos covering agent memory, subscription tracking, and multi-agent collaboration. The through-line: open source is making tools that used to cost hundreds a month, or require a data-science team, both free and customizable.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
An agent skill that turns a spoken or typed system description into an interactive, editable architecture diagram (45.6k stars) instead of a static generated image.
01:57 – 04:12
02 · God's Eye View
A browser-based console pulling together live planes, ships, satellites, wildfires, and public street cameras into one navigable interface; hosts debate real use cases like flight and cargo-ship tracking.
04:12 – 05:33
03 · OpenMAIC
Type a topic and get a full AI classroom back, slides, an AI teacher, and AI classmates, positioned as more immersive than static YouTube tutorials or chat-only tools.
05:33 – 06:36
04 · Zapier SDK
Sponsor segment: Zapier's SDK exposes its roughly 9,000 app integrations to a builder's own software with one connection instead of building each integration by hand.
06:36 – 07:48
05 · Go Modern Guidelines
A JetBrains plugin that feeds a coding agent current Go packages and language features released after its training cutoff, so it stops recommending five-year-old patterns.
07:48 – 10:21
06 · OpenClaude
An unauthorized fork of Claude Code rewired to run on OpenAI, Gemini, DeepSeek, or local models; its own license file admits it lacks Anthropic's authorization, and both hosts couldn't find a reason to use it over the already-open OpenCode.
10:21 – 11:42
07 · Scientific Agent Skills
Over 160 narrow, specific skills (genomics databases, chemistry calculations, paper drafting) that turn a general AI assistant into a working scientist's tool for one exact niche task.
11:42 – 13:21
08 · MiniMind
A complete, readable PyTorch recipe for training a small language model from scratch (pre-training through LoRA and DPO) on regular hardware for about the price of a coffee.
13:21 – 15:45
09 · OpenSEO
An open-source rebuild of the $129/month SEMrush/Ahrefs SEO category, free software with a $10/month hosted data source; one host signs up live on air.
15:45 – 18:18
10 · Omarchy
DHH's opinionated, AI-native Linux distribution that revives old laptops and treats an AI agent as a first-class part of the desktop; named after 'omakase' dining.
18:18 – 20:15
11 · TimesFM
Google Research's time-series forecasting model, version 3.0, which projects future numbers (clicks, signups, temperature) without a data-science team; the hosts already use it internally.
20:15 – 21:27
12 · Hindsight
First audience submission: a coding agent skill that reviews its own past session and encodes lessons so the agent, and the developer, stop repeating the same mistakes.
21:27 – 21:45
13 · Model Deck
A macOS menu-bar app showing how much of each AI subscription's usage is left, broken down by project, session, and task.
21:45 – 22:30
14 · Cortex Suite
An AI memory system and codebase analyzer built to make long agent sessions more reliable, with an integration for people already using Graphify.
22:30 – 23:06
15 · RTA Smriti Brain
A local-first, evidence-backed project-memory tool for coding agents named after the Sanskrit word for memory, so a new session doesn't have to reconstruct project reality from scratch.
23:06 – 23:51
16 · YouTube Subscriptions Ingest
A first-time GitHub contributor's tool that pulls YouTube subscriptions into structured local Markdown using Andrej Karpathy's LLM-wiki framework; the first ingest processed 11,500 videos from 41 channels in ten minutes.
23:51 – 25:21
17 · Reimagine It
A redesign skill that reads an existing HTML file as the design brief and applies its actual styling to new components, instead of inventing a generic look from a text description.
25:21 – 27:14
18 · Radiant
A native Mac app where several AI agents share one conversation and consult each other mid-task, reachable from a phone, signs in with subscriptions instead of API keys, and can run local models with no key at all.
Atomic Insights
Lines worth screenshotting.
Archify turns a spoken or typed system description into a clickable, editable architecture diagram instead of a static generated image, because engineers communicate better with a whiteboard than with words.
God's Eye View is a single browser interface for plane, ship, satellite, wildfire, and public street-camera data that already exists scattered across the internet.
OpenMAIC turns any topic into a full AI classroom with slides and AI classmates, addressing the gap between static YouTube tutorials and chat-only tools like ChatGPT.
Zapier's SDK exposes roughly 9,000 app integrations to a builder's own software with one connection, instead of each developer building every integration from scratch.
AI coding models are trained on a fixed knowledge cutoff, so a Go plugin like go-modern-guidelines exists purely to feed the agent packages and features released after that cutoff.
OpenClaude is an unauthorized fork of Claude Code rewired to run on OpenAI, Gemini, DeepSeek, or local models, and its own license file states it lacks Anthropic's authorization to distribute the proprietary source.
Both hosts couldn't identify a concrete reason to use OpenClaude over the already-open-source, already-free OpenCode, and recommended contributing to an existing open tool instead of forking a proprietary one.
Scientific-agent-skills is deliberately narrow rather than generic: most of its 160+ skills target a single specific database or lab task a working scientist would recognize by name.
MiniMind's pitch is training a real language model end-to-end (pre-training, fine-tuning, LoRA, DPO) on regular hardware for roughly the price of a coffee, in readable PyTorch.
OpenSEO rebuilds a $129/month SEO category (SEMrush, Ahrefs) as free open-source software, charging only $10/month to host the underlying data source instead of licensing the software itself.
The advantage of an open-source SEO tool isn't just price, it's that people can build custom CRMs and dashboards on top of it that a closed SaaS product would never let them touch.
Omarchy is a full Linux distribution built by DHH (Ruby on Rails' creator) where an AI agent is a first-class part of the desktop, not just a chatbot bolted onto macOS or Windows.
Omarchy takes its name from 'omakase', the Japanese restaurant tradition of the chef choosing what you eat, because DHH pre-selected an opinionated set of tools and lets you strip out what you don't want.
Google's TimesFM projects any time-series data (clicks, temperature, signups) forward without requiring a data-science team, and the hosts already use it internally to project course signups.
A first-time GitHub contributor's YouTube-subscriptions tool ingested 11,500 videos from 41 channels into searchable local Markdown in ten minutes, four weeks after they first learned GitHub.
Hindsight's core lesson for coding agents applies to humans too: when something breaks three times, stop patching the symptom and go find the upstream cause.
A submitted memory tool for coding agents (Cortex Suite) integrates with Graphify specifically so an agent doesn't have to rediscover a codebase's structure every new session.
Several submitted repos (Hindsight, RTA Smriti Brain, Cortex Suite) all attack the same problem from different angles: a new AI coding session shouldn't have to reconstruct what a past session already learned.
Takeaway
Open source is quietly replacing expensive, closed AI and dev tools.
WHAT TO LEARN
Across ten trending repos and seven reader submissions, the same pattern repeats: something that used to cost hundreds a month or require a specialist team is now free, self-hostable, and built by one person.
01Archify
Words alone don't communicate system architecture well, which is why AI diagram tools like Archify are replacing 'describe it in chat' with a clickable, editable visual output.
02God's Eye View
A tool doesn't need to be a new category to be useful, it just needs to make scattered, already-public data (planes, ships, satellites, cameras) navigable from one interface.
03OpenMAIC
Static video tutorials and pure-chat AI both have a gap: neither creates an immersive, interactive learning environment, which is the specific niche a tool like OpenMAIC is filling.
04Zapier SDK
Before building custom integrations for every tool your users need, check whether a service like Zapier's SDK already exposes thousands of them through one connection.
05Go Modern Guidelines
An AI coding agent's knowledge is frozen at its training cutoff, so plugins that feed it current package and language information are a cheap fix for a permanent problem.
06OpenClaude
Before adopting a fork of a well-known tool, check whether it has a stated reason to exist beyond 'run this on different models' when a free, open alternative already does that.
Read a repo's own license file before assuming it's authorized. OpenClaude's license admits in writing that it lacks the rights to distribute the proprietary code it forks.
07Scientific Agent Skills
A tool aimed at a narrow, specific audience (scientists needing one exact database skill) can be more valuable than a broad, generic tool nobody quite needs.
08MiniMind
You don't need a research lab to train a language model. A complete, readable recipe covering pre-training through fine-tuning can run on regular hardware for the price of a coffee.
Small, custom-trained language models built for one narrow job (writing your email subject lines, for example) are coming, and tools that make training them approachable are worth watching now.
09OpenSEO
The real advantage of open-sourcing an expensive SaaS category isn't just the lower price, it's that people can build custom dashboards and integrations a closed vendor would never allow.
Bloat and lack of customization are the actual openings a scrappy open-source alternative needs to attack an established, expensive incumbent.
10Omarchy
An operating system doesn't have to be treated as fixed. If it gets in the way of how you want to work with AI, an agent-native rebuild of the OS itself is now a real option.
Borrowing a curation philosophy (like 'omakase', letting an expert pre-select what you get) can be a legitimate product design choice, not a limitation, as long as the choices stay removable.
11TimesFM
Time-series forecasting used to require a data-science team; feeding historical numbers into a model like TimesFM and projecting forward is now accessible to anyone tracking clicks, signups, or any recurring metric.
12Hindsight
A coding agent that reviews its own past session and encodes what went wrong is solving the same problem good engineers already know: don't just fix the bug, fix why it happened.
When you hit the same failure three times, stop patching the symptom and trace it upstream to the actual cause, this applies to debugging code and to any recurring personal mistake.
13Model Deck
Tracking exactly how much of a recurring subscription's usage is left, rather than guessing, is a small but real utility gap most AI subscription products still don't fill themselves.
14Cortex Suite
Multiple people are independently building the same idea (persistent, evidence-backed memory for AI coding agents) right now, which is a strong signal that it's a real, unmet need rather than a novelty.
15RTA Smriti Brain
An evidence-backed record of a project's current reality, kept between sessions, saves an agent from re-deriving the same context and the same conclusions every single time it starts fresh.
16YouTube Subscriptions Ingest
A brand-new GitHub contributor with four weeks of tutorials behind them can still ship something that processes 11,500 items in ten minutes. Experience level isn't the gate to shipping something useful.
17Reimagine It
A redesign tool that reads your actual existing HTML as the brief produces more consistent, 'yours' looking output than one that has to guess your taste from a text description.
18Radiant
If you want an audience-submitted project to get a strong reaction, ship a clear README with screenshots. Two hosts here spent thirty minutes just trying to understand ones that didn't have any.
Glossary
Terms worth knowing.
MCP (Model Context Protocol)
A standard connection an AI agent uses to call external tools and apps through one integration point instead of building a custom connector for each one.
LoRA
A lightweight fine-tuning technique that adapts a trained language model to a new task without retraining all of its parameters, making it cheap enough to run on modest hardware.
DPO (Direct Preference Optimization)
A method for tuning a language model's behavior directly from examples of preferred versus non-preferred outputs, used as an alternative to more complex reinforcement learning setups.
Agent skill
A packaged, drop-in instruction set that teaches an AI coding assistant how to perform a specific real-world task, such as running a chemistry calculation or drawing a system diagram.
Omakase
A Japanese dining style where the chef selects the dishes for you; Omarchy borrows the name because its creator pre-selected an opinionated toolset for the operating system.
Time-series forecasting
Predicting future values of numerical data that changes over time, such as website clicks or temperature, based on its past pattern.
“It's me. It's AI. It's everything. It wants pictures. It craves screenshots. It craves graphs like this.”
Personifies AI's hunger for visual context in one punchy exchange, works with zero setup.→ TikTok hook↗ Tweet quote
09:09
“I don't know if we need another open source version of Claude Code. We've got some of those, you know, maybe try and support one that's already out there.”
Blunt on-air critique of a submitted-adjacent tool, rare candor for a repo roundup show.→ IG reel cold open↗ Tweet quote
10:08
“OpenClaude's own license file states in writing, this project does not have Anthropic's authorization to distribute their proprietary source.”
The legal-tension line, makes a strong headline clip about the fork's own admission.→ newsletter pull-quote↗ Tweet quote
15:05
“I'm so excited about this that I'm signing up for it while you're talking.”
Unscripted, live reaction, captures genuine enthusiasm for a $10/month tool replacing a $129/month one.→ TikTok hook↗ Tweet quote
17:00
“I keep trying to shave ounces off of the code.”
Tight racing metaphor for DHH's minimalist engineering ethos on Omarchy.→ newsletter pull-quote↗ Tweet quote
19:59
“If you're just doing basic averages right now, this is going to get you probably 40% more accurate.”
Concrete, testable number claim about TimesFM's forecasting lift.→ TikTok hook↗ Tweet quote
20:31
“When something breaks three times, don't fix it, go upstream and see what caused the issue and fix that.”
Standalone actionable rule that reads as advice for developers and non-developers alike.→ newsletter pull-quote↗ Tweet quote
23:28
“The first ingest did 11,500 videos from 41 channels in 10 minutes.”
Eye-popping throughput stat from a contributor who learned GitHub four weeks earlier.→ IG reel cold open↗ Tweet quote
The Script
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
17px
metaphoranalogy
You're going to get a gorgeous visual that will explain your project's code easily. You'll have an insane way of turning an old dead computer into a futuristic AI device. It is amazing by someone you know.
You're going to get a stunning way to spy on the world. Is it a little creepy? Well, you're going to see some real life use cases for it.
That's this week's top 10 GitHub repos. Links to everything below, including to my co -host LinkedIn, if you want to connect with him there. Timestamps so you get what you need.
Let's get into it. Presented by Zapier, the AI automation company. Adam, the hottest repo of the week is called Archify.
And what it does is it describes your system chat in a clickable diagram. Let me show you what it looks like. And I'm wondering what someone would use this for.
Yeah, I mean, this is a great way. You know, we often will describe with words. to AI what we want to build, but words don't really capture relationships and flow and directionality.
And so when engineers are typically building a new system, I've got a whiteboard over here. I draw it on the whiteboard first because that's a way better way to communicate. And I'll take a picture of that and I'll send it to AI.
What this is, is a tool that helps you create these system architecture diagrams with AI. So you don't have to get the whiteboard out. You don't have to get dirty.
And then you can edit them. You can watch as it changes and treat it as like documentation and a spec for what you want to build. And this looks like a tool that's like a little more than some of the diagram tools we've seen so far that are just generated image.
This one's actually interactive. You can click and drag. You can move things around and edit too.
I'm a visual person, too. OK, I could totally understand how AI is a visual person, too. Andrew, it's you.
It's me. It's AI. It's everything.
It wants pictures. It craves screenshots. It craves graphs like this.
I guarantee if you use something like this, your designs, your systems, your software will come out much more accurate with a lot less work. Okay, I can see that. I can see how I would understand it better this way.
Next is something just super cool. We got to show a video. It's called God's Eye View.
It's a spy satellite console in your browser, and it uses real data. I mean, there's a lot of this stuff that's already live on the internet, like live planes, ships, satellite information, wildfires, public street cameras. But how do you get access to all of them?
How do you see it all? What this repo did is it said, okay, I understand that this stuff exists. I'm going to make it easy for you to navigate.
In fact, look, this... This gif here is a really good visualization of how it makes all of that data accessible. Really cool interface.
I don't exactly know what the use case is. I've got this video here for people who are downloading this file where you can see it. In fact, here, let me just do a quick hit play on this.
much more. He keeps showing the White House for some reason. This seems a little suspicious.
But what would I even use this for? You know, Andrew, I was just thinking that same question. The one thing that pops up for me is, you know, number one, like I'm sure you use FlightAware or one of these other tools to track like, oh, my my spouse is coming home.
I'm going to pick her up from the airport. I'm going to track it. I don't want to use the crummy like app from the airline.
And so I use FlightAware or something. But. The second idea that came into my head, I've shipped things internationally a lot before on these big shipping vessels, and they don't have tracking.
They'll be on the water for four weeks, but the boat is tracked. And so if you can figure out what boat it's on, and I've done this before, but it's hard to find, you know, one of these platforms that lets you in for free and say, oh, I know which boat. Oh, now I can track it.
I know it's it's almost to Panama. It's almost you know, it's it's in the canal. It's moving through.
So I think this is kind of interesting. There's a lot of like tracking needs that we have. And I want status updates.
I don't know about. All the, yeah, the picture of the White House and the camera views. I don't know what all that's for, but I like the plane and boat tracking.
I could see how I would use this. I moved to Mexico. My wife wants to go to a new city here.
I ask Claude. I ask friends. It's not the same as just seeing Street View, for example, and seeing what else is going on.
All right, makes sense. Let's move on to the next one. I don't know how to pronounce this one.
It's called OpenMake, I think. But here's what it is. You know how when you want to learn a topic in the past, you'd go to YouTube.
Today, you're going into like ChatGPT. And the problem with YouTube. is it's just consistent and static the problem with chat gpt is yes it's interactive but it's all chat what this does is it creates basically an online education like class for you with fake students with slides with an immersive experience that's what it's about here's this uh it's a paid tweet here i want to call that out but he did a really good job of explaining what it is and here and you can see also what open make stands for it's open massive AI empowerment course and if you go to the repo they do actually have a sample here where you can see what it's like to learn but all you have to do is say I want to learn about whatever and it will create this classroom for you to learn I see the value in this one this one's really interesting I mean, this is like all those skills we've seen previously where you could call a skill to say, teach me this topic and it'll create a PDF, but far more interactive and in an interface.
Is that is that right? Yeah, exactly. Exactly.
All the things that you used to. I'm going to not let it load right now and leave it to people to download this after we're done. By the way, if you are building, you should know about my sponsor Zapier, which makes.
They're SDK available to you. Here's why, Adam, people would use this. And I say Zapier because when you hear Zapier, you go, oh, I have all this image of what it used to be.
No, there's so much more today. So there are people who are listening to us right now who are building software for outsiders. They're building software for their employees at a company that they're working for.
And what they want to do is take the software and make it. enable their users to connect it to the other tools that they use, to their project management software, to their email software, et cetera. But to build those connections is painful.
So Zapier SDK makes it really easy. You just use Zapier SDK and then you say, this is what I want my users to have access to. I want them to use my software with Notion.
I want them to use my software with et cetera over, at this point, it's almost 9 ,000 apps, probably even more than that. It keeps growing. That's the value of this.
Go tell them that you heard about it. For me, they were sponsoring this before they even knew what this was. All right, Zapier SDK.
Let's talk about this. Go modern guidelines, JetBrains fix for agents that write five -year -old Go. I've heard of Go.
It's a language, but what is this? Yeah, Go is a programming language. It's a backend programming language.
So it runs on servers typically. Think of it as like similar to Python, somewhere between Python and C or C++. And, you know, a common frustration with things that change a lot with AI is my AI model was trained in November, and now it's September.
Well, wait, all the things that have developed, the new releases, the feature changes in the language, none of that is in the model. And so I'm getting stale recommendations. And so this, and there's a lot of tools like this, you know, if you program in different languages or have a different topic, go and find one that's for you.
This will plug in. to Cloud Code or Codex so that Codex is aware of, oh, here's the best package that was released three months ago. You got to use this one.
And here's how you have to use it. Here's the examples. The LLM doesn't know, but it could quickly learn.
And this is the tool to make it quickly learn those things in an automatic way. All right. Can I say, by the way, I'm so proud of these graphs that I've been building.
It's working. I save all the stars every week and I'm able to build it. Okay.
Next, OpenClaude. Claude code's workflow pointed at any model you want. This is a fork of Anthropic's Claude code rewired to run on OpenAI, Gemini, DeepSeq, or a free model on your own laptop.
It is not from Anthropic. And its own readme file says so in writing. And we'll talk about the controversy with that.
What do you think? What do I need to know? So this is OpenClaude.
I'm familiar with OpenCode and then obviously CloudCode. You know, CloudCode is not an open, you know, not open source, but OpenCode is open source. It's been around for, I would say, a long time now in the AI era.
Like, why would I use this instead of OpenCode that's already open? It's already free. It's well supported.
It has a big community around it. I'm trying to find, like, what is... What is the reason why I would use this?
And you and I spent some time before we got started trying to find that and we just couldn't find it. I don't get it. You know, one thing I would point out, if you're out there and you've been using open or sorry, cloud code, you're like, oh, I want to find a way to use this cheaper or free.
Go and try Codex actually, too, because I think of all of these open code is a great. open free, you know, platform. But Codex can also use all these other models.
It's open source, so you can fork it if you want and change it. And the interface in the CLI is far better than than cloud code. So try some of the other tools like rather than this is maybe my recommendation to GitLab is is maybe try some of the other tools.
Like, I don't know if we need another open source version of cloud code. We've got some of those, you know, maybe try and support one that's already out there. Or I would love if someone could come back and say, no, open cloud is better in this one very specific way.
Because sometimes some of these tools, they're very opinionated about one thing and they solve a problem that nobody else does. I wasn't able to figure out what that was from reading about it. I'm going to send this to the GitLab, the GitLab people, GitLab people, and see if they can fill me in on this.
By the way, legality of this is unsettled. The repo says it itself. OpenClaude's own license file states in writing, this project does not have Anthropix authorization to distribute their proprietary source.
Users and contributors should evaluate their own legal position. I'm surprised they would even use the name Claude, considering what happened to Claude. I mean, yeah, that doesn't go well.
So we'll see what happens here. Scientific agent skills. This is a set of over 160 skills that turn your AI assistant into a working scientist.
Just drop this instruction pack and it teaches an everyday AI assistant real lab work, query and genomics database, run the chemistry calculation, draft the paper, et cetera, instead of being walked through every step of the way. You like this one and usually hate when their skills on the top 10 list for the week. Why are you not hating on this one?
You know, I hate generic skills. Here's how you'll write better. Here's how you'll, I don't know what, but this is not a set of generic skills.
This, most of these are wildly specific to hit this here, hit the NCATS ARACs. I have no idea what that is, but I'm sure. One scientist out there is like, oh, I know exactly what NCATS ARAX is.
And here's why I use it. And here's what I have. And now they have a skill that's going to help them do their work.
And they're maybe not as AI savvy as everybody else. And so they can have this that's just going to make whatever their weird specific niche science thing was. That's going to get easier as a result of going into these skills.
This is not a package where a scientist would go in and use all hundred and how many ever 163 skills. People are going to go in. They're going to find.
the one or two things that are perfect, custom built for them, and they're going to go. So I love, I love that. I love the depth there.
All right. That's the number six hot repo of the week. Number seven trending is called mini mind.
Build your own language model for about the price of coffee. The Eli five is the complete free recipe for training. Any tiny language model from absolute scratch, pre -training, fine tuning, Laura, DPO with every algorithm written in plain.
How do I pronounce this? Type pie torch. PyTwitch?
Anytime you see Python or Py, P -Y, it's P -I -E, like the pie you eat. So you can read what is actually happening. You like this one.
Why? What is it? I love it, man.
I mean, small language models are coming. They're not here yet, but... This time next year, definitely two years from now, every person, every business is going to have a bunch of custom trained small language models to write your email subject lines or your email, whatever the thing is that you're doing, you should have a custom model that's working for you.
And in order to get to that world, we're going to need a lot of tools just like this one. to help average people train models. Right now, that's pretty hard to do.
The tools are hard to understand and the big labs are capable of it. But very few people are building easy to understand tools like this that can run on regular hardware. So I absolutely support this one.
I love it. I'm definitely going to go and try it this week. Needle is a small language model, right?
That we covered. Needle is a small language. Yeah, yeah.
But that one was like a pre -built model. So you could think I could build a bunch of custom needles by using ManyMind. I got it.
Okay. All right. By the way, the founder of Needle saw that he was on.
He's going to come on and explain to me how to use Needle and give me all kinds of use cases. So cool. Coming up, number eight.
I know this one really well. Open SEO. Here's what it is.
I think this title needs a little explanation. The $129 a month SEO tool rebuilt as a pay -per -search. the 129 reference is actually even understated it's a reference to how much you pay for semrush and ahrefs when you need software to help you with your seo so what this guy ben decided to do was he said i think we can open source this and make it free what you do need though adam is a data source something to tell you what's connecting to what what's on the internet that data source costs And so he goes, look, I'll give the software for free.
People can sign up for the data source for pennies or they could pay me 10 bucks a month and I'll host this software for them and get them the data included, which is peanuts. The beauty of this is not just the money. We keep talking about how open source and GitHub gets you free versions of all these tools.
The beauty of this is the customization. Adam, I saw people create CRMs that connect into open SEO. So if they're asking for backlinks, the CRM keeps track of it.
Beautiful dashboards for people who are not into SEO to keep track of how their stuff is performing in search engines. And so, yes, SEMrush and Ahrefs have a big advantage in the fact that they have... built their software over the years.
But for many people, it's too bloated. And more importantly, you can't customize it. That's what Ben built here.
We have a whole show on that. It's fantastic. People love it.
I mean, I'm really excited. I'm so excited about this that I'm signing up for it while you're talking. I mean, we use all those other tools.
They're really expensive. They're really slow. I want to add another seat.
And they say, well, that's another thousand dollars a month or some crazy thing. So I'm so excited. I think if this can deliver what it's promising, I'm over the moon.
I'm going to sign up today and find out more. In fact, a lot of people did. After I interviewed Ben and did the video, he sent me an email.
He goes, Andrew, we got so many customers because of this. Actually, so many users. He didn't say that anyone necessarily paid him, but he doesn't seem to care.
I don't know how his business is going to survive when he's charging only $10 a month for hosting. But all right, this is a start and people are loving it. Maybe if, you know, hook him up with us, we might be able to help him find a better business model for it because clearly there's a lot of opportunity in this space.
And a lot of love. All right. Yeah.
I'll make the introduction to GatewayX, the venture studio in St. Louis. Okay, I have been so wrong about this in so many ways and I'm so obsessed with this.
One way that I was wrong is I called it Omachi, like a lot of people on YouTube. DHH calls it Omachi. It's like, what is it called?
Omakasa, where you go into a Japanese restaurant and they give you the thing that they think you want, right? That's where the name comes from. What this is, is Linux.
that you can run on an old laptop and the AI agent is part of the experience and an AI agent that's not just Siri, but it's more than that, Adam. And this is what's exciting me about it. People are saying, I don't like how my operating system lives or works.
I don't want to have to accept it. With Mac, I have to accept it, or maybe I find some tool that I can add that will jank up the system and make it do the thing I want. Here, I can just tell my AI, rewrite the freaking operating system and give me what I want.
And that kind of customization, that kind of built -in AI is really interesting, and it's really powerful. And more than that, now... I think what you're also getting is DHH's taste.
And yes, it is kind of omakase. He's, for example, added a video editor onto this and a video shorts editor into this. Even though he says, look, I want this to be like a Formula One car.
You know, he's a racer. And I like how they will shave ounces off of their car. And he goes, I keep trying to shave ounces off of the code.
He still is willing to. put some software on there that he thinks you need and want. And yes, you can remove it if you want it.
Really beautifully done. People have taken old laptops, started to use this. Suddenly Linux is really hot because of this.
And our boy Alex Finn, who's excited about everything, is super excited. Let's just say, I got to hit play on this. Alex always like...
The thing that just came out is always going to change the world. The entire internet is going nuts talking about this brand new agentic operating system named Omachi. And for good reason.
It is the only AI native operating system I've ever used. And it has made it basically impossible for me to go back to Mac OS or Windows. Until Mac OS comes out with a new laptop next week and he gets excited about that, you need to switch to Omachi immediately.
I don't know if I share his definitive enthusiasm, but I'm excited about it. What do you think about this? Yeah, you know, I...
Mac often gets in the way of my AI agent. I'm trying to build something. I'm dealing with security and it won't let me install this.
And now I've got to go and do a bunch of things. Having an operating system that's intentionally meant to be used and even changed by agents is a pretty compelling, pretty compelling promise. All right.
Next, final one for the week. And then we're going to go into user submissions, viewer submissions. It's Google's forecasting model.
It's called Times FM. And it's the new version that just came out. What is it?
And why do we even care? Yeah, there's a lot of what's called time series data in the world. You know, you've got, say, your SEO stuff we were just looking at.
Well, how many clicks did you get? How many impressions per month, per week? And as you look at all of these, you know, even the temperature.
Well, how do you predict what the temperature is going to be next, right? If I'm watching it this morning, well, there's a lot of modeling software out there and you can hire an engineer. This aims to make it very easy to take any time series numerical data.
and project out into the future what it might be next, right? If you gave this the last week's worth of temperature data and you see how it goes up during the day and down at night, this model would do whatever machine learning type stuff has to be done. You don't have to learn how to do it.
It will take all of that and it will project out. You say, oh, you want another week's worth? Great, it'll give you another week's worth of estimated temperature data.
We use this, we actually use this exact package internally to help project statistics for, you know, one of our e -commerce brands. to project what signups will be for a course that we offer. And, you know, it's just better than the basic math we were doing before, which was how many have we gotten, how much time, and then here's how many we expect in the next amount of time.
So it's like a little bit better estimations. You know, if you're just doing basic averages right now, this is going to get you probably 40 % more accurate. And this is from Google Research.
It is not someone who's stealing Google's name. All right. Next, first stars.
Adam, we've had people who will email me afterwards and say, you featured me. I now have 30 stars, 40 stars. It is amazing.
On GitHub, people who are brand new to this are submitting what they've done, often for the first time. I want to show some of it. I won't be able to get to all of it.
We'll save it all for, we'll save the rest for next week. But let me show a few here. This one is called Hindsight.
It's a coding agent that reviews its own session. What do you think of this? what is this i think it's really cool uh if you haven't seen it there's built into cloud code a slash command called i think it's insight or insights that also will do something like this and i think there's a lot of how do you stop making the same mistakes, right?
And how do you, as you start new coding sessions, how do you learn from your past mistakes and encode those changes so that you don't make them again? And it's not just your mistakes, it's the agent's mistakes too. So something like this is there to help you just improve, improve, improve.
That way you're not burning tokens, you get to a better result faster. By the way, one thing that I've learned from you is take these repos and give it to my agent and say, based on how I work or based on my skills that I've created, what can I do? I have a feedback loop that at the end of working with my agent, I will then go through a feedback loop so that I can improve it.
And I said, how would I use this and what he's done here to improve my approach? And it gave me six. tips based on what he did.
And one of them is when something breaks three times, don't fix it, go upstream and see what caused the issue and fix that. I like that. And I'm going to find ways to integrate this into my feedback loop.
Next is something called model deck. See how much your AI subscriptions. you have actually left.
This is pretty. I'm going to show a screenshot and then we can move on because you can tell immediately what this is. It shows you how much of your agent's usage you still have left on your subscription.
Really sweet. Like it a lot. Next, Cortex Suite AI memory system and code base analyzer.
The memory system and code base analyzer built to make long agent sessions more reliable with an integration for people already using Graphify. What is it? Or what do you know about it?
Yeah, it's like one of those memory systems that says, well, I can look at everything that's there and I'm going to give access to the agent, the design of your system so that the agent can make smarter choices and that this will also last between agent sessions. So the agent doesn't constantly have to rediscover, oh, where was that file?
Where was that function? What's the structure of the database? This kind of keeps all of that in one tool layer.
two stars i am the second when you see this viewer go and see how many stars by the time this comes up okay next is rta smrita smriti brain he explained the name of it and what it comes from here it is the name isn't an acronym it's a cosmic order truth in In the rig Vita smear tea is memory. Okay project memory So a new session doesn't start from scratch local first evidence -aware project memory for AI coding agents It keeps an evidence -backed view of a project's current rely reality.
So the next session doesn't have to reconstruct it and I think that explains it really well. I'm going to leave it there.
It's got seven stars, including one for me. Let's go to the next one. This is great.
YouTube subscriptions ingest. 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 Andre Karpathy's LLM wiki framework, raw data into a local staging folder processed into structured markdown with a prompt asking whether you want local.
to cut the token overhead. Basically, it allows you to study YouTube videos quickly and use them and use the data from them in an organized markdown, right? Yeah, I love that.
Oh, look at this. Wait, I got to read this. The first ingest did 11 ,500 videos from 41 channels in 10 minutes.
Okay. Next, reimagine it. A redesign skill that reads your file instead of inventing a vibe.
You looked at this. What can you tell me about what it is? Yeah, I mean, scroll down to see some pictures.
First of all, I love the aesthetic that this brings. And anytime you can get an aesthetic into your coding agent that's different than the default, you're going to start to see outputs that are uniquely you and that are special. So I love that.
This, though, is about... How do you create a design system in a way that the agent's going to understand? So you get to sit here, you get to fiddle, you get to play with a bunch of different traits of how do I want it to look?
And then once you have that, you save your HTML file that encodes all of that styling. And now the agent can just copy and paste those styles into whatever you say. Oh, I want a button over here.
Great. It'll copy and paste that style. It'll copy and paste the header.
So it kind of is speaking the language that the agent wants to speak versus you describing. I want a purple box that's floating that looks really pretty, blah, blah, blah. But that's not really detailed enough to give the agent what it needs.
So I love this. I'm going to try it. Like I said, I love the aesthetic too.
So I even just enjoyed scrolling around. Like I'm one of those weird people, Andrew, that's got like art books hanging around. Like this is a book of logos and stuff.
Like I just, I love to scroll through this to get inspiration. So I highly recommend it. I do too.
I want more of that. More inspiration because then it gives me ideas and I say to my agent, now copy this and make it mine. That's exactly right.
Feed it into my system and help me adjust it. Okay. I think we're going to end with this one.
It's called Radiant Multiple AI Agents in One Shared Conversation. It's a native Mac app where several agents share one conversation and consult each other mid -task. Backed by a server you can reach from your phone.
Signs in with subscriptions. I love that. Instead of API keys and runs local models with no key at all.
It's got two stars, so it's very early. Actually, I thought I gave it a third. There we go.
I will say this. This could really use a few screenshots to help us see how this works. Hopefully, by the time that you see this viewer, it will be up there.
If you are new, you should know some of these people did not have their repos online, did not have it on GitHub public until we went back and forth by email. And then they said, OK, screw it, Andrew, I'm going to show it. And then things started to go.
Andrew, we should put together a little guide. We've looked at so many of these. We should put together a guide for people of here's how you should lay out your readme.
Here's what your screenshot should show. You know, we've looked at so many and some are awesome, but they're hard to understand. So I'd actually like to help people describe these things better.
So when people submit them to you, it's like, oh, this is cool. And we maybe even help them grow with that as well. I would like that.
It would make my job a lot easier. Sometimes when someone sends me one of these repos, it takes me like half an hour. I get sucked in trying to understand it and trying to figure it out.
And I think if they did a better job with the readme, it would help. Okay, again, this is so brand new. We're talking about two stars and three stars now.
So I don't mean to be critical. I just mean to be helpful. I'm glad that it's there.
If you've been watching and you haven't subscribed, please subscribe. It's a good way to show us. If there's one of these that has been helpful for you, let us know in the comments.
If there's a different type of show that you would like, know in the comments, too, because you've seen me. I'm now doing these roundups so you can get a few different repos around a specific topic, but I'm fishing for topics.
So say all that in the comments. You've got my email address. And oh, now that this is over, I've got another set of 10 that you really need to see.
There's a link for you right here. We'll see you there.
The Hook
The bait, then the rug-pull.
The cold open stacks four teaser lines, a gorgeous auto-generated diagram, an old laptop turned into a futuristic AI device, a browser console for spying on the world, before either host says a single repo's name out loud.
CTA Breakdown
How they asked for the click.
VERBAL ASK
26:54subscribe
“If you've been watching and you haven't subscribed, please subscribe. It's a good way to show us. If there's one of these that has been helpful for you, let us know in the comments.”
Soft, single ask at the very end, paired with a request for topic suggestions in the comments and a plug for the resources link and submission email shown on screen throughout the audience segment.
Add Modern Creator as a preferred source and Google shows you more of our breakdowns in Search, Top Stories, and AI Overviews. It only changes what you see, and you can undo it in your Google settings anytime.
Add to Preferred SourcesOpens your Google source preferences with us pre-loaded. Tick the box and you're done.
Two hosts count down the week's ten hottest GitHub repos, from a security-scanned Claude plugin index to a self-hosted finance tracker wired to 2,500 banks.
Corey Ganim and Andrew Warner speed-run a week of AI launches, an AI QVC host, an endless AI livestream, and a phone number for your agent, and call out OpenClaw 2.0's rough demo along the way.
A weekly roundup episode where the host and a Zapier guest run through ten open-source repos built to pair with a coding agent, from screenshot-to-code to a proxy that cuts terminal noise by up to 90 percent.
Ben Senescu built OpenSEO to kill his own Semrush bill, wired it straight into Claude, and now watches open-source contributors fork it into CRMs and content tools he never planned to build.
00:24
00:58
01:45
02:15
02:59
03:31
03:58
04:42
05:11
05:18
06:04
06:25
07:03
07:21
07:39
08:26
08:51
09:17
09:42
10:08
10:51
11:11
11:31
11:51
12:31
12:51
13:11
13:57
14:21
14:45
15:09
15:33
16:23
16:48
17:39
18:05
18:47
19:06
19:26
19:45
20:05
20:42
21:00
21:18
21:36
21:42
22:07
22:22
22:48
23:00
23:28
23:43
24:24
24:37
25:09
25:49
26:08
26:26
26:45
27:04
18:45
07:04
16:34
21:02
13:57
19:02