←
pdp13,059 posts
pdp avatar

pdp

@pdp

Building autonomous AI agents. Receipts: - usenoodle.app - chatbotkit.com - cbk.ai - zot.im

◷ Joined February 2008▧ 1,853 Photos and videos
pdp avatar
pdp@pdp·2h

Interesting

🎉 Introducing SiliconForest: a new open operating system built on Darwin and XNU, the open-source core of macOS, running natively on an M1 MacBook Pro. Here’s the interesting part: it runs unmodified Mac apps. This first version runs Apple’s Terminal and Activity Monitor, copied straight from a macOS install, plus the stock Mac build of Firefox. Working today on a 13" M1 MacBook Pro: GPU acceleration (Vulkan 1.4), hardware video decoding, Wi-Fi, sound, keyboard and trackpad, and the Touch Bar. It’s Apple’s open-source kernel, with the parts Apple doesn’t publish reimplemented from scratch. I’ve also brought my Wayland compositor, Chonkstep, over to Darwin, and the results so far are fantastic. The idea: a free, hackable OS that can run native Mac apps. It’s early. Big apps like Logic Pro and Photoshop need a lot more (Metal, for a start), but that’s the direction. It stands on great work: PureDarwin (which we fork), the Asahi Linux GPU and hardware drivers, Omacom Mesa, and the Finch frameworks project. Source is coming soon. Right now it runs on one machine, the 13" M1 MacBook Pro. If you’d like to help bring it to other M-series Macs, get in touch. github.com/chriskearney/sili…

1:10
pdp avatar
pdp@pdp·3h

👀

Today we disclose CVE-2026-123633 and CVE-2026-123007 (CVSS 4.0: 9.2) - exploitable memory corruptions in NGINX We achieved reliable remote code execution (ASLR defeated) on NGINX serving static files - no backend or third-party modules 🤯 Both require HTTP/2 Read more 🧵⬇️

0:22
0:33
pdp avatar
pdp@pdp·5h

It is a tale as old as time. Developers are pretty bad at estimating effort. That is why we have product and project managers. If you decide to build your own PostHog: 1. You will spend as much on LLMs as the PostHog team. They use AI too, and they have bigger budgets. 2. Your solution will be limited in scope because you have one customer. You. 3. You miss out on network effects, like bugs and security issues reported by other customers. 4. The LLM writes the code, but you still provide the intent, the testing and everything else. 5. Security and all the rest is on you, with your limited resources. Sometimes it makes sense, especially when the margins are not justified. But often it ends up as expensive as the subscription you were trying to avoid, if not more.

Full disclosure: I used to work at PostHog I know most of these posts are clickgagement but for the genuine folk who end up believing that building in-house or self-hosting is the better option... I can guarantee you PostHog is the most economically viable option - Their free tiers are insanely generous, not to mention their startup plan - Believe it or not there are still economies of scale in SaaS, not to mention increased reliability of a managed solution - The have one of the best billing control options (second only to AWS that I've seen) - They literally offer to refund accidents: posthog.com/handbook/growth/… (only co. I've ever seen do that) For anyone who legitimately burns through $$$ of PostHog credits I would goes as far as to say that you should refocus your efforts on internal spend, I'd guess there's serious improvements that could be made to your own bottom line

pdp avatar
pdp@pdp·5h

My question is how quickly organisation can adopt to rapid patching practices.

GPT-6 Cyber for helping further secure the internet, with a major security flaw in Java found by Calif, as part of our Patch the Planet initiative: nytimes.com/2026/11/17/us/po…

pdp avatar
pdp@pdp·5h

I think this is a good approach and it makes sense. The only challenge is to make this practical and useful. By constraining the execution environment you get more security and observability at the expense of more rigidness.

Giving an AI agent access to a terminal is convenient. Giving it unrestricted access to the host machine is a different decision. A terminal can expose files, processes, credentials, network access and system configuration. That is a significant security boundary. An interesting open-source project in this space is Bashkit. GitHub: github.com/everruns/bashkit Bashkit is a Rust-based virtual Bash interpreter designed for sandboxed execution. Its approach is particularly interesting because commands run in-process rather than spawning ordinary host processes through fork and exec. It also provides a virtual filesystem, resource limits, optional network access and script analysis. This gives an agent platform more control over what a script can do before execution. The broader architectural idea is that an agent should not need unrestricted operating-system access simply because its interface resembles a terminal. A purpose-built execution environment can provide familiar developer capabilities while limiting the consequences of unexpected commands. For anyone building coding agents or automation platforms, Bashkit is worth exploring. #OpenSource #Rust #AIAgents

pdp avatar
pdp@pdp·5h

I am really surprised by the powers of Swift.

pdp avatar
pdp@pdp·6h

This is what I love about Noodle. I know exactly how much my agents are spending.

Find out which bot is hoarding your disk.

0:20
pdp avatar
pdp@pdp·6h

I might be able to make CISOs' job a lot easier - at least when it comes to getting real-time information.

pdp avatar
pdp@pdp·6h

There are 5 well known AI hacking tools and just about 0 defensive ones. That's about to change.

pdp avatar
pdp@pdp·6h

Grok Bot but for cyber security defence 👀

Post attachment
pdp avatar
pdp@pdp·7h

I am looking for 3 companies willing to pilot Pion - AI-powered defensive security tool - first of its kind. It will be a close collaboration to make sure that the tool works for you and your teams. Anyone interested?

pdp avatar
pdp@pdp·8h

You can just build things like this with Noodle because why not?

Post attachmentPost attachmentPost attachment
pdp avatar
pdp@pdp·19h

I am tempted to just: defaults write com.pdparchitect.noodle NoodlePacks -array security and not bother with another tool...

pdp avatar
pdp@pdp·19h

It turns out there is a better way.

pdp avatar
pdp@pdp·20h

For now I think that it makes more sense to run your agentic defensive tools (ADT?) locally on device rather than a centralised system because there is risk it could be either part of the attack or you may end up not having access to it during an incident.