Pangolin HA, Cloudflare AI Scraping, and a Portable Homelab

Homelab Highlights for 2026-09-16: self-hosting and homelab notes worth knowing.

[Audio embed placeholder]

News Roundup

Nonprofit that tracks meteors taken down by “critical blow” from a cyberattack

The International Meteor Organization, a nonprofit coordinating meteor observations since 1950, suffered a cyberattack that dealt a critical blow to its aging infrastructure. The attack has taken much of its site offline, and the group expects several weeks of partial downtime as it transitions to new services.

Why it matters: It’s a sobering case study for anyone maintaining critical services with limited resources, highlighting the vulnerability of volunteer-run infrastructure and the operational impact of such attacks.

Source: Security - Ars Technica

Cloudflare Updates AI Crawler Controls, Defaults Raise Questions

Cloudflare changed its recommended setting for AI Training crawlers from ‘Block’ to ‘Disallow AI Training,’ which allows search indexing but instructs crawlers not to use content for AI model training. However, users checking their dashboards found the actual default setting was ‘Allow,’ prompting discussions about control over content scraping.

Why it matters: For anyone publishing content online, this highlights the ongoing challenge of controlling how data is used by AI companies and underscores the importance of actively managing web application security settings.

Source: blog.cloudflare.com

Community Homelab: Downsizing a Full Lab for a Cross-Country Move

A homelab owner packed up a substantial setup—including Threadripper workstations and multiple high-end GPUs—for a move and is now running on a compact, temporary lab built from NUC-style machines. The downsized setup handles media serving and AI inference, with power draw dropping dramatically from 1-4 kilowatts to 100-600 watts.

Why it matters: It’s a practical demonstration of efficiency gains through hardware and software optimization, challenging the assumption that serious homelab work requires rack-mounted, power-hungry hardware.

Source: claude.ai

Plexamp Native Linux Build Now on Flathub Beta

Plexamp’s native Linux audio engine build is now available on Flathub’s beta channel for x86_64 and arm64 architectures. This marks the first time the application is available from Flathub for ARM-based devices like Raspberry Pi, offering better performance and lower resource usage than the older Electron-based version.

Why it matters: For users of Plex media servers on Linux, especially those with smaller devices, this provides a more efficient, officially supported way to run the dedicated Plex music player.

Source: dl.flathub.org

OpenHome Foundation Companion App Roadmap Proposal

A core iOS application engineer has submitted a detailed proposal for a more native companion iOS application for the OpenHome Foundation’s home automation platform. The proposal is open for community feedback on the project’s public roadmap.

Why it matters: For users invested in the OpenHome ecosystem, this signals potential improvements to the mobile experience and demonstrates the project’s open development process for major features.

Source: github.com

Tool & Software Highlight: Pangolin 1.23: High Availability and Clustering for VPN and Reverse Proxy

Pangolin 1.23 introduces integrated high availability clustering for self-hosters, removing the need for external proprietary components. The update streamlines multi-node deployments for its identity-aware remote access platform, which handles VPN and reverse proxy duties. High availability features are included in the Enterprise Edition, which is free for personal use with a license key.

Why it matters: This brings enterprise-grade redundancy and failover capabilities to self-hosted remote access and application publishing, a significant step up for homelabs and small teams that require reliable uptime.

Source: github.com

Sources

Transcript

Host A: Welcome to Homelab Highlights, the show that surfaces practical homelab wins and useful self-hosted tools.

Host A: Pangolin 1.23 just shipped, and the big news is proper high availability clustering for self-hosters. This is the identity-aware remote access platform that handles VPN and reverse proxy duties. Until now, you had to use external proprietary components to run multiple nodes, which made deployment cumbersome unless you used their cloud service.

Host B: Version 1.23 changes that by integrating DNS resolution and certificate management directly into Pangolin itself. According to the release notes, this makes setting up a highly available system much more streamlined. The high availability features are included in the Enterprise Edition, which is free for personal use with a license key.

Host A: There’s also a consolidation happening with their tooling. Newt, the tunnel connector that runs on your remote network, is now integrated directly into the Pangolin CLI. So you can start a site with the same tool you use for SSH and client connections. Existing Newt deployments keep working, but the direction is clear: they’re moving everything into the CLI.

Host B: [thoughtful] The community reaction notes some interesting tradeoffs. One commenter pointed out that burying certificate management inside Pangolin makes them nervous—they prefer running Traefik plus ACME separately to see every cert issuance. And the shared state across nodes relies on PostgreSQL, so you still need to manage a highly available database cluster yourself. That’s where HA setups often get tricky for self-hosters.

Host A: Moving to the feed, the International Meteor Organization—the nonprofit that coordinates amateur and professional meteor observations—suffered what they’re calling a critical blow from a cyberattack. Their aging infrastructure took much of their site offline, and they expect several weeks of partial downtime while they transition to new services. For a volunteer-run group that’s been around since 1950, that’s a massive undertaking.

Host B: They’re prioritizing fireball observation reporting during the outage, but it’s a sobering reminder that even niche nonprofits running critical infrastructure can be targets. The article doesn’t specify the attack vector, but aging infrastructure is mentioned as part of the problem. It’s a case study for anyone maintaining services with limited resources.

Host A: Next, Cloudflare made changes to their AI crawler controls. According to their blog post and customer emails, the recommended setting for AI Training crawlers shifted from Block to Disallow AI Training as of September 15th. Here’s what that means: Disallow AI Training lets major search crawlers like Googlebot index your site for search results, but tells them not to use your content for training AI models. Allow, on the other hand, permits both indexing and training.

Host B: But here’s where it gets interesting. Users checking their dashboards found the actual default setting was Allow, not Disallow. So if you read their email saying most customers won’t need to change anything, you might want to double-check your security settings. The underlying issue of controlling AI scraping is real for anyone publishing content.

Host A: Switching gears to hardware, there’s a great community post about downsizing a homelab during a move. Someone packed up their permanent lab—which included Threadripper workstations and multiple RTX 3090s—and is now running on just a few NUC-style machines with an RTX 5090 on an eGPU dock connected over oculink. The temporary setup handles everything from media serving to AI inference, proving you don’t always need rack-mounted hardware for serious work.

Host B: They even designed a custom enclosure with Claude Fable and FreeCAD when the stock enclosure had terrible thermal management. The big benefit: power draw drops from 1-4 kilowatts down to 100-600 watts. That’s the kind of efficiency gain that makes you reconsider whether you need all that hardware running 24/7.

Host A: Quick note on Plexamp: the Linux Flatpak is now on Flathub’s beta channel for both x86_64 and arm64. That means Raspberry Pi or Asahi Linux machines can run it from Flathub for the first time. The new version uses Plexamp’s native audio engine instead of Electron, which means better performance and lower resource usage on those smaller devices.

Host B: And finally, a clever hardware project: an ESP32 POE-powered LLDP reader. This little device tests for POE power, then reads LLDP data from the switch to display the hostname, management IP, and port number on an OLED screen. Unplug it and everything clears. It’s designed for port tracing without needing expensive network testers, and the total build cost is under $50.

Host A: Let’s dig into that ESP32 POE LLDP reader project because it’s exactly the kind of practical tool that makes homelab life easier. This is an open-source project on GitHub that builds a portable device for network troubleshooting. Plug it into a network port—if it powers on, you’ve got POE. Then it reads LLDP data from the switch and displays the switch hostname, management IP, and the actual port number on a small OLED screen.

Host B: What makes this clever is the temporary nature of it. Unplug the device and all collected data is cleared by design—no stored credentials or network information to worry about. The creator used heat shrink during assembly and field testing, so it’s meant to be rugged. For someone managing a rack with unlabeled cables or troubleshooting connectivity issues, this beats carrying around a Fluke or similar expensive tester.

Host A: [quick aside] There’s an important warning in the documentation: the ESP32-POE board must not be plugged into USB power while using POE. There’s an isolated version of the board that can handle both, but for this use case, you wouldn’t need it. The project page has all the details, including parts list and assembly instructions. It’s one of those builds that solves a specific problem elegantly without over-engineering.

Host B: For our community highlight, let’s return to that downsized homelab story because it’s such a relatable scenario. When their spouse got a job across the country, they packed up what was essentially a small data center: Threadripper workstations, multiple high-end GPUs, a Synology NAS with 48TB RAID—the whole nine yards.

Host A: The temporary setup is three compact machines: a Ryzen 7 mini-PC with an RTX 5090 on an eGPU dock, a Ryzen 5 machine, and an Intel N150. They’re running Ubuntu, and the key insight is that Qwen3.8-27B handles 95% of their LLM use cases. That’s the kind of software optimization that often gets overlooked when we focus on hardware specs.

Host B: [with emphasis] The thermal management fix is particularly interesting. The stock ORIGIMAGIC A1 enclosure had poor cooling, so they disassembled it, took measurements, and used Claude Fable to design a custom enclosure with a Noctua 120mm fan in FreeCAD. That’s the homelab spirit right there—not just accepting hardware limitations, but engineering around them. They’re even considering keeping this downsized setup permanently and selling the heavier workstations once unpacked. The power savings alone make that tempting.

Host A: That’s Homelab Highlights for Wednesday. Until next time, happy hosting!