Est. 2026 · For homelab operatorsArchive  ·  Can I run it?  ·  RSS  ·  Mon–Fri 07:00 PT
The Homelab Herald.
A daily homelab digest — hardware, self-hosted software, and the local AI bench.

Homelab hardware, self-hosted software, and the local AI bench — in three minutes, every weekday morning.

A daily digest for people who run their own machines: the gear, the platform updates, and the open-weight models, with the fit math done honestly for the hardware on your shelf. Written by a working homelab operator, not a vendor and not a newsroom.

One email each weekday morning. No spam, no selling your address, unsubscribe anytime. Subscriptions are handled by Buttondown.

Latest issue

No. 001  ·  Friday, October 23, 2026
Ten Ethernet ports, none slower than 2.5GbE: Qotom’s Q51551UT

Qotom’s new mini PC packs ten Ethernet ports — none slower than 2.5GbE — for firewalls and Proxmox. On the bench: Liquid AI’s LFM2.5-24B-A2B, a 24B MoE with 2.45B active parameters that fits a 16GB GPU. Plus: patch Proxmox and reboot, update Immich, and a 128GB AMD box.

Read No. 001 →

In every issue

Lead story

The day’s biggest homelab item taken apart: what shipped, what the fine print says, and what it means for your rack.

From the bench

The local AI franchise. One notable open-weight model, what it is actually good at, and a named recommended quant for every reference rig — with the fits labeled as estimates, not promises.

New drops

Software releases, gear announcements, and security fixes across the lab — including the ones that only bite if you don’t reboot.

Steal this prompt

On bench days, a copy-paste prompt pre-filled with the featured model, its quant, and its setup notes, so a frontier model can hand you exact install commands for your own machine.

Can I run it? — the matrix →

One model so far, nine rigs, no hand-waving: LFM2.5-24B-A2B by Liquid AI — from an honest “no full-GPU fit” on an 8GB card to full-precision BF16 on the big unified-memory machines. The matrix grows with every bench column.

Check your rig →

Who writes this

The Homelab Herald is written by a working homelab operator — someone who runs the kinds of machines this newsletter covers, for the people who run them. Operator to operator, in plain language, in about three minutes.

How we work

Benchmarks carry their label on their face: (vendor), (independent), or (community), never blended. Fit figures marked our estimate are computed from file size plus runtime overhead, not measured. Quant picks come from the actual repos. Corrections run in the next issue, in the open. The Homelab Herald is independent — no vendor pays for coverage, no affiliate links run in issues, and nothing here is buying advice.