retiring the pigeon homelab

Today I’m retiring my beloved Raspberry Pi homelab, hostname pigeon. I bought it back in 2023 after the stock of Pis finally recovered from the COVID-19 pandemic. It’s always functioned as a homelab, perfectly quiet and low-power, tucked away under a shelf on my desk, crunching away at the bits and bytes of my digital life.

It hosts a handful of Docker-based applications including a services directory (Madness), an RSS feed aggregator (FreshRSS), a YouTube archiver (Pinchflat), a file server (NextCloud), and an Obsidian notes server (CouchDB). These services help me exercise sovereignty over my data and preserve my privacy against the prying eyes of corporate monoliths. They empower me to be aware of and choose where my news comes from, to maintain copies of my favorite videos even if the platform deems them unsavory, and to own my digital files rather than renting space for them on someone else’s computer.

However, there are still some areas where the Pi’s modest specs limited its utility. I tried to spin up a backup solution for my photo library (Immich) and personal movie collection (Plex), but media processing pipelines proved too demanding. Instead, I’ve been paying for iCloud storage for my photos and using the Home Sharing feature on my Mac mini to stream to my Apple TV. And despite my reservations about the rising prevalence of large language models (LLMs) like ChatGPT, they are becoming increasingly capable and I’d like to give them a fair chance. Thus, I need more powerful hardware to evaluate the viability of hosting open-weight models like GPT-OSS and Qwen, so I can leverage the power of this emergent technology without sacrificing my privacy.

To bridge the performance gap and expand the range of services offered by my lab, I’ll be migrating over to the Albatross. This PC build has a beefier Intel i9-12100K CPU, NVIDIA RTX 3060 GPU, and 32 GB of RAM. I’ve migrated my NextCloud instance, and I plan to extend it with a Google Docs-like web editing experience via ONLYOFFICE. I’ve also spun up an LLM chatbot application with OpenWebUI that can run larger models (30B parameters) with good throughput (up to 50 tokens/sec). I’m going to give Immich another shot and hopefully cancel that iCloud subscription. Although I’ll likely continue using Home Sharing on my Mac mini to stream movies, since the files remain on my device.

As for the Pigeon, retirement isn’t the end of the line but the start of a new chapter. Maybe it’ll find its new role as the brains behind a digital photo frame, the orchestrator of a smart home bridge, or the server for a silly little website, possibly this one. Only time will tell.