Sam: An entire AI team. Sixteen specialist agents. All running locally, on your desktop.
Kai: I'm Kai.
Sam: And I'm Sam.
Kai: And you're listening to Trending Repos for Monday, November 2nd, 2026.
Sam: So, Kai, what have you got for us?
Kai: We're starting with an AI agent army for your computer. It's called Orkas, and it promises to turn your high-level goals into finished tasks.
Sam: Then, a project for anyone who's ever squinted at a `git diff` and wished for more. OmniDiff is a new difftool in Rust that's content-aware for images, PDFs, you name it.
Kai: And we'll wrap with a new way to find your next great technical read: HackerBooks, which scrapes over a decade of Hacker News comments for the most-mentioned books.
Kai: Alright, let's get into Orkas-AI/Orkas. Sam, this feels like a huge leap: an MIT-licensed, local-first AI team that lives on your machine.
Sam: 'Team' is a strong word, Kai. It’s a collection of scripts. And 'on your machine' immediately brings up questions about resource usage, privacy, and security.
Kai: But the architecture is what's so cool! You give a 'Commander' agent your goal, it creates a to-do list, and then delegates to sixteen built-in specialist agents.
Sam: Sixteen agents? What are they specialists in?
Kai: All sorts of things! There's a file manager, a code writer, a web researcher... and it can even coordinate with external models like Claude Code and Codex for the heavy lifting.
Sam: Hang on. It's 'local-first' but it's calling out to external APIs like Claude? That sounds a lot less local than advertised.
Kai: The coordination is local! The 'brain' is on your desktop. Think of it like a manager calling outside contractors to augment the local team.
Sam: An augmented team that's still running arbitrary code and accessing my file system. I'm not saying it isn't powerful, I'm saying it's a massive attack surface.
Kai: Okay, but imagine what you could build with this. The repo is clear: you download the release, give the Commander a goal, and it just gets to work. No complex setup.
Sam: A goal like what? 'Make my coffee'?
Kai: Not yet. More like: 'Analyze the user feedback in this folder, propose three new features, write the tech specs, and draft a blog post announcing it all.'
Sam: Right. I'm just saying, before you tell it to 'refactor the entire codebase for performance,' maybe double-check your backups. And your API keys.
Sam: Alright, next up is something a bit more grounded. A new open source difftool called OmniDiff, from GitHub user ivankovic.
Kai: I saw this on Hacker News! Written in Rust, so you know it's going to be ridiculously fast.
Sam: It is fast. But speed isn't the main event. It's 'content-aware.' The creator actually researched what an 'ultimate diff tool' would look like by studying diffs in open-source projects.
Kai: Content-aware how? It knows syntax?
Sam: It knows syntax for 25 languages, yeah. But it also knows how to diff images, PDFs, zip archives, even cursor files.
Kai: Wait—it can diff a PDF? How? Does it just spit out a wall of binary garbage?
Sam: Nope, it renders them side-by-side or just shows you the text changes. Same for images. It gives you a meaningful comparison where `git diff` just says 'Binary files differ.'
Kai: That is a complete game-changer for designers or anyone working with visual assets in a repo. So how do I use it? Do I just replace my git command?
Sam: Not quite. You just configure it as your Git difftool. So when you type `git diff`, it launches the OmniDiff graphical UI instead of the default one.
Kai: I am so tired of pulling down two versions of a PNG just to see what changed. I'm setting this up the second we're done recording.
Sam: It's a very sharp tool for a very annoying problem. I'm impressed.
Kai: Alright, to wrap things up, let's talk books. This is HackerBooks, at hackerbooks.org. It's inspired by Andrej Karpathy's HackerTube, but for books on Hacker News.
Sam: Okay, a recommendation engine based on HN comments. What's the dataset look like?
Kai: It's huge. The site says it has about 16,500 books mentioned by at least two different people, going all the way back to 2007.
Sam: Sixteen thousand books... so what's the signal here? Is a book good, or just controversial enough to get talked about?
Kai: And get this—the creator says they had Claude Opus build the whole thing. The parsing, the site... everything.
Sam: Built by an LLM? Okay, now I'm really questioning the data's integrity. And again, is 'mentioned twice on Hacker News' really a seal of quality?
Kai: I think you're missing the point. It’s not a 'best books' list, it’s a cultural index. And here's the key: every book links directly to the comments where it was mentioned.
Sam: Ah. Okay, that changes things. Context is everything. You're not just seeing the title, you're seeing why people brought it up. That's actually useful.
Kai: Exactly! So if you want to find the most influential systems design books, according to HN engineers, you can see the recommendations—and the debates—in real time.
Sam: I could also use it to find the most over-hyped pop-sci books to make fun of. So, win-win.
Kai: There you go! Something for everyone.
Kai: Alright, quick recap. We started with Orkas, the local AI agent team that automates complex tasks on your desktop.
Sam: Then OmniDiff, a super-fast difftool that understands everything from code to PDFs and images.
Kai: And we finished with HackerBooks, a browsable index of every book making waves on Hacker News for the last decade and a half.
Sam: So Kai, before we go... speaking of AI generating things, you know what month it is, right?
Kai: Oh no. Don't say it.
Sam: It's November. Which means it's NaNoGenMo: National Novel Generation Month. The challenge to generate a 50,000-word novel, entirely with code you write yourself.
Kai: Maybe I can delegate that to my new Orkas agent team.
Sam: Just don't give them your AWS keys. And that's our show! I'm Sam.
Kai: And I'm Kai. We'll be back tomorrow with more Trending Repos.
This show is made with AI: the hosts’ voices are synthetic and the scripts are AI-assisted. Every story links to its original source.