I let AI deal with my Plex transcode and saved 8.5TB with out lifting a finger

My Plex library had grown far too giant, and it was time to downsize it. I might delete stuff, certain, however I needed to maintain most of what was there. So, I turned to changing my whole library to H.265, and it saved me practically 10TB of house.

Downsizing my NAS meant downsizing my library

Deleting outdated motion pictures and TV reveals solely received me thus far

I not too long ago made the transfer from running my old enterprise-grade NAS with 12 drive bays and 80TB of storage to a UGREEN NAS with 6 drive bays populated with three drives and a complete of 40TB of storage.

This downsize saves me fairly actually lots of of {dollars} monthly in electrical energy, nevertheless it did not come with out its downsides. My Plex library is quite sizable, measuring at nicely over 40TB earlier than the transfer.

So, I began off by deleting issues that no one had watched in years. I’ve been working my Plex library for about 6 years at this level, and a few content material had been on there because the starting and by no means watched.

I exploit Codex for every part today, and I had it combine with Tautulli to seek out the media no one had been watching and provides me a listing of delete candidates. A similar thing could be accomplished with Maintainerr, I simply selected to do it with Codex.

From there, I nuked tens of terabytes of media from my server that I simply actually did not “want.” I by no means did it earlier than as a result of I had 80TB of usable house, so I simply stored every part.

Once I deleted every part I did not want, I used to be all the way down to about 30TB. That may sound like quite a bit, contemplating I had 40TB to work with, nevertheless it nonetheless wasn’t snug sufficient for me. So, I made a decision to take the subsequent step: transcoding my whole Plex library.

The final time I attempted transcoding it took perpetually and was a ton of labor

I had Gemini writing my ffmpeg scripts and it wasn’t fairly

In early 2026, I spent some time transcoding some of my movies from H.264 to H.265. It was a ache within the neck and a whole mess, to be sincere. I undoubtedly received the job accomplished, nevertheless it was difficult.

I used to be utilizing AI to assist me give you the ffmpeg scripts, nevertheless it was all handbook. I’d feed my wants and setup to Gemini (the AI I used to be utilizing on the time) and it might give me a script to make use of. I’d check it till it failed, then attempt to debug it with Gemini.

There was loads of back-and-forth right here, and that was probably the most painful a part of all. Things would seem like they had been working effective, after which fail in a single day and waste hours upon hours of time that would have been spent transcoding.

Overall, it labored nevertheless it did not work how I needed. I used to be working with a large library and solely regained 7TB of house.

This time round although, once I began with proper round 30TB of media, I went a distinct route—I had Codex deal with it from begin to end.

Shrinking my library this time was easy and simple

H.264 to H.265 was straightforward once I let Codex do the heavy lifting

One of my favourite options of Codex is that it could possibly run scheduled duties. So, once I got down to transcode my library this time, I made a decision handy the duty off to Codex.

Codex used SSH to configure all of my nodes that I’d use for Tdarr, in addition to constructing customized ffmpeg scripts and plugin flows inside Tdarr for every node. In complete, I ran 5 nodes to get by the transcoding as quick as doable.

I selected to transcode to H.265 as a result of my Plex server now runs on a more moderen platform that is able to dealing with HEVC content material natively. Plus, H.265 takes up significantly much less house than H.264.

Some of the nodes used NVIDIA CUDA, whereas others used built-in Intel graphics. This was one of many largest issues once I arrange Tdarr the primary time, because the plugin movement was a ache. Codex dealt with it like a champ although.

I principally instructed Codex what I needed and it configured every part for me. From there, I instructed Codex to watch the transcode standing each quarter-hour. So, each quarter-hour, it might replace the thread with the present progress and in addition flag any errors.

Here’s the place Codex began to essentially pull its weight—I had it repair the errors routinely for me. I wasn’t nervous about shedding media at this level, I had a backup on my outdated NAS and I might re-rip something I wanted to.

So, through the 15-minute checks, if Codex noticed a problem, it might repair the script and re-queue no matter failed. I principally let Codex run the transcode for about 72 hours, as that is how lengthy it took for every part to course of by.

At the tip of all of it, I regained 8.5TB of house simply from Codex transcoding my media. Overall, I shrank my media library from round 40TB to round 16TB, which left me with loads of headroom for the server to develop over time once more.

AI has confirmed its usefulness in my homelab time and time once more

While AI is much from good, it is undoubtedly nice at an excellent many issues. Having it run this transcode was like having an IT professional sitting in my workplace monitoring it each quarter-hour with out taking a break for sleep or meals. I undoubtedly could not have achieved this with out Codex’s assist, a minimum of, not within the time I did it in and with as little frustration as I had.