A NAS starts out as a big project, but once its storage pools are set up and all services are running smoothly, it’s just a box sitting quietly in the corner doing its job. Mine had started to blend in with the furniture, and the only time I thought about it was when I wanted to host a new service or a hard drive failed.
The machine that houses my most important data, running Plex for my entire familyand managing my backups didn’t have a way to let me know how those tasks were going. If you wanted to know, you would have to analyze the log files or click on a web dashboard to find a meaningful status report. I realized my life would be a lot easier if I could ask my NAS what’s taking up so much disk space or how the backup job went last night.
My server is a repurposed gaming rig that was awesome a decade ago. It can’t run the latest titles anymore, but it’s exactly enough to run a small local LLM. I installed OllamaIt pointed out some things that were already running on the machine, and now the most important box in my network can tell me what’s going on inside it.
My NAS can finally respond
I needed a way to interpret the data I already had.
Part of why this setup works so well is that the local LLM lives on the same machine about what I’m asking about. For example, when I ask it something about my Plex library, it doesn’t need to use an API or query a remote service as it simply works with local files. That simplicity plays a big role in why the setup is still viable on older hardware.
Ollama handles the LLM side. Setting it up is as simple as installing the program, pulling a model, and starting the service to start listening on localhost. I’m running Llama 3, which admittedly isn’t very good, but it’s sufficient for this setup and is low on resources, which is perfect for an older system. Open WebUI sits in front of it as the browser-based chat interface, meaning I can open it from any device on my local network.
Most applications write log files or can be configured to do so. In my setup, TrueNAS has its own log files and Plex keeps a detailed log of playback history. Since these are just plain text files, they can be searched with tools already on the system. A short Python script reads the logs, passes the content to the Ollama API, and is interpreted by the local LLM.
What you can really ask
Read the text you would rather not read yourself.
What is all this? The configuration basically boils down to a log summary.. TrueNAS is verbose, so things like SMART checks, cleanups, and failed mount attempts are written in detail and not easy to read at a glance. Asking the model to summarize the last 24 hours is one of the most useful ways to get the most out of it. It will read all the logs and flag anything that looks like a warning, making it easier to find the details you’re really interested in.
I also like to use the model to learn more about Plex. Plex stores its logs in an SQLite database, which is just a file on disk. My Python script dumps its contents to output the playback logs to a text file. From there, I can ask the local model what my friends have been watching this week or if they’ve seen anything new that I haven’t seen. It will read the dump, analyze the titles and usernames, and give me a quick breakdown of how everyone has used Plex during the week.
This won’t work well on Synology or a low power ARM box
Local AI simply uses too much RAM and CPU for some systems
If your NAS was built just to host files and do almost nothing else, or if your system has between 4 and 8 GB of RAM, then this setup won’t be feasible. Even an older, lighter model like the Llama 3 needs a decent amount of RAM. I think a lot of home labs have converted their old gaming rigs into NAS machines, as I did, and they’re the ones who will be attracted to this setup.
since I am only ask model to read log files and tell me if you find anything worrying, it doesn’t put that much pressure on the hardware. Complex tasks and encoding prompts would give very different results, and those aren’t the kind of things a NAS should be used for anyway. Since no one is impatiently waiting for last night’s SMART results, getting slower responses isn’t much of a trade-off, but it still takes reasonable system specs to work.
The server that joins your network should be smarter
I really feel like my NAS now has a brain. Instead of just storing everything, you can give me details about your operations and alert me to any problems that appear in the logs. If you’re already running a decent x86 server, this costs nothing except a little extra RAM. Ollama, Open WebUI, and a simple Python script are free and only take an afternoon to get up and running.
