Who needs Netflix when you have Plex, the media server software that runs locally and lets you stream your own local video files? I’ve been hosting my own Plex server on a Raspberry Pi 5 for a few years now, and it’s mostly run without a hitch.
But like taking your car to a mechanic even when nothing’s obviously wrong with it, you should periodically look under Plex’s hood to see if it needs the equivalent of an oil change. The thing is, I’m no expert when it comes to streaming video. I know (roughly) what an MKV video container is, and I know how to organize my Plex media libraries, but I’m clueless when it comes to the ins and outs of video transcoding.
In short, I need help, and—you guessed it—I turned to AI, and specifically I turned to Gemini. (I could have turned to ChatGPT or Claude, too, but for the moment I’m a Google AI Pro subscriber.)
Like any other tool, LLM models like Gemini are good at some things and terrible at others. A chainsaw is great for pruning a tree, but it’s not the right tool for cutting a wedding cake. By the same token, LLMs tend to suck at life coaching and genuine creative writing (they’d rather just plagiarize), but they excel at analyzing reams of error logs, and for a Plex checkup, that’s precisely the skill that’s needed.
Now, there are a few ways to let an LLM pick over your Plex installation. One way is to tell it your basic setup (Windows, Mac, or Linux) and let it feed you terminal commands to pluck out the correct log files, which you then paste into the chatbox. It works, and it’s relatively safe, since you’re not actually letting ChatGPT, Claude, or Gemini loose in your system. But it’s also slow and tedious—your CTRL-C and CTRL-V fingers will quickly tire. (If you do prefer sticking with an AI chatbox, I have a custom GPT system prompt for you below.)
Instead, I chose to go cutting-edge. I’ve been playing with Google Antigravity, the Google-made IDE (integrated development environment) that lets Gemini and other LLMs (including Claude) do the coding for you, guided by your natural-language prompts. Yep, we’re talking vibe coding.
Google Antigravity is just one such AI-driven IDE. OpenAI has Codex (including the recently released Codex for macOS), and Claude Code and Cursor are other examples. With Antigravity, you can give Gemini access to a “workspace” on the target system (generally one or more directories that you’ve designated), and with your guidance, it will read files within the workspace and can even—gulp! — execute terminal commands. Gemini can also think about what it wants to do and present you with an “implementation plan” before actually doing it.
The degree of Gemini’s autonomy in Antigravity is up to you; for example, you can have it request your approval before every terminal command, or if you’re willing to take your hands off the wheel, there’s an “always allow” setting that lets Gemini go to town, executing shell commands at will. Google recommends the safer “always approve” setting, and I tend to agree.
Anyway, back to Plex. I’d previously set up a Plex workspace on my Raspberry Pi for Antigravity’s use. (I use a specific subdirectory as a staging area complete with “git” version control, perfect for retracing your coding steps in case of a foul-up.) I fired up Antigravity, opened my Plex workspace, and gave Gemini 3 Flash (a faster and cheaper “regular” version of Gemini, versus the pricier high-grade Gemini Pro 3 model) a prompt: Can you give my Plex server a check-up?
Further reading: Plex’s lifetime subscription plan is getting a massive price hike
Gemini went quiet for a moment (“Thinking…”) and came back with a plan. It would check my Raspberry Pi’s vitals—CPU temperature and load, memory usage, storage availability—as well as check a half-dozen Plex logs, poke at the Plex database, test-drive my server’s transcoding and networking performance, and then ensure my external storage drives were properly mounted and delivering decent data throughput. Essentially, we’re talking a full-on Plex physical.
After looking over the plan, I clicked the blue Proceed button, and off Gemini went, snippets of terminal commands flying around as I occasionally clicked the “Approve” button.
(This is a good time to note that in doing this analysis, I was giving Gemini full access to all my Plex media files. Needless to say, you’ll want to carefully consider whether you’re comfortable with that prospect.)
About five minutes later, Gemini came back with its diagnosis, and my eyes popped open at the results. It reported that my Plex server itself was A-OK (“peak athletic condition!” it said, leaning into the check-up metaphor); but like a doctor spotting a worrying test result during a routine exam, Gemini’s Plex check had unearthed a critical system issue: evidence of a failing SD storage card.
Now, this surprised me because I’d already—or so I thought—migrated from booting my Raspberry Pi from that fragile SD card to a much more robust NVMe storage module. But Gemini found scores of I/O errors originating from the old SD card that I thought I’d replaced. It’s initial hypothesis: my Raspberry Pi might still be using the flimsy SD card for bootup tasks.
Um, say what? Had I completely forgotten to migrate essential system files from the SD card even after investing in a new NVMe and spending a weekend installing it? (Groan.)
Like any worried patient, I sought out a second opinion. In a new Antigravity instance, I summoned the big guns—Claude Opus 4.5—and dumped Gemini 3 Flash’s prognosis into the chatbox. “Take a look at this Plex checkup report, is it true,” I asked?
Clause dove in (“Thinking,” again), peeked into more system logs, and came back with a reassuring answer: No, my Raspberry Pi wasn’t booting off the old SD Card. But it was still checking the card since I’d left it plugged into my system, and those checks were clogging my logs with I/O errors. In other words, Gemini saw yellow flags in the Pi’s bloodwork and jumped to a dire—and lucky for me, incorrect—conclusion.
Armed with this knowledge, I took the details back to Gemini and asked it to reassess. It soon came back; yes, my Pi was booting off the NVMe drive, as expected, and yes, the still-connected SD card was the source of all the I/O errors.
Gemini’s overall Plex diagnosis? B+, it said (I had asked for a letter grade). Vitals look good, the plumbing’s in nice shape, but that old SD card is like a splinter in the Pi’s toe, it said. Yank it at the first opportunity. Oh, and maybe clear Plex’s transcoder cache in case of any stray buffering.
So, there you go: a recipe for regular AI-guided Plex checkups, perfect for spotting silent Plex server ailments. Give it a shot—you might discover something unexpected, too.
And if you want to try a Plex server once-over using a custom GPT in a chatbox, here’s that system prompt I promised—and yes, it’s courtesy of Gemini itself, prompted by me:
This story is part of TechHive’s in-depth coverage of the best media-streaming devices.
Author: Ben Patterson
Source: PCWorld
Reviewed By: Editorial Team
