USB has always been a no go, works to a point then collapses, in my experience (buying every high end Alfa USB card I could get my hands on).
I used to do that. But then I found OpenWRT which took away all the grunt-work and just worked OOB. So why wouldn't I use that instead?
For bonus points, block it from accessing the internet itself.
I just need to get my controller setup now, so I can play with the dashboard. I had heard some features are missing from the controller that are on the AP web interface though. Which of them had you noticed missing?
(should probably disclose that I am one of the devs that maintained the apu2 coreboot boot firmware)
My problem with APs running proprietary FW is that I don't trust them to be secure, even if the vendor does updates, you never know what they're doing in the background. E.g. some APs have a hidden secondary SSID for their proprietary mesh implementation. With OpenWRT I can set them up exactly the way I want to, using open standards (mesh, roaming) instead of vendor-specific crap.
I'm happy with it, but I did have to get a heatsink for it, since otherwise it overheats easily. Since I got it they released a couple of dual-band dual-concurrent cards like this one: https://www.asiarf.com/shop/wifi-wlan/wifi_mini_pcie/wifi6e-... , which is pretty neat, since you don't need to get a separate card for 2.4GHz devices.
One more thing - how is the signal range for you?
> One more thing - how is the signal range for you?
With four 5dBi antennas it's sufficient to have >800mbps in every corner of my single bedroom apartment. Other than that I have no means to test, sorry :)
On a semi-related note I will say it's infuriating that replacement Linux networking stack components are released with features missing. I hadn't dug into nftables enough yet to say whether that is the case, but netplan and, to a lesser degree, systemd-networkd have driven me bonkers.
Most of my gripes were related to IPv6 and DHCPv6-PD. The people supporting the replacements never seem to be in any rush to add missing features back into the replacement. Most would think you'd wait until the replacement has reached feature parity with it's predecessor(s) before pushing it out to mainline.