Homebox Companion - AI-powered photo cataloging for your Homebox

[Duelion]

Hello everyone!

I've been working on an unofficial companion app that adds AI-powered item detection to Homebox. The idea is simple: take photos of your items, and GPT vision automatically identifies and catalogs them for you: names, descriptions, quantities, labels, and more.

Quick feature highlights:

• 📸 Snap photos, AI detects and catalogs items automatically
• 🏷️ Multi-image analysis for better accuracy
• ⚙️ Customizable AI behavior (configure how fields are generated)
• 🐳 Docker deployment ready
• 📱 Mobile-friendly web interface with QR code scanning

It's still early days, but it's been really helpful for quickly cataloging large batches of items without the manual data entry grind. Thought some of you might find it useful too, especially for bulk inventorying rooms, shelves, or collections.

demo_video_10mb.mp4

I've set up an online demo that uses your demo server as the target Homebox instance. It's currently running on a limited spending OpenAI key for testing purposes, which I'll eventually revoke.

The app authenticates with your existing Homebox credentials and integrates directly with your location tree, labels, and API. Everything runs as a separate service, so it doesn't require any modifications to Homebox itself. (BTW - great work on the Homebox API!)

Please let me know if I'm stepping out of line by using your demo server for this. I was considering adding this demo link to my repository for others to test, but wanted to check with you first.

Check it out: https://github.com/Duelion/homebox-companion

Would love to hear your thoughts or feedback if anyone gives it a try!

[Saturo78]

First of all, I want to say a big thank you for the effort you’re putting into the Homebox companion app. It’s great to see the ecosystem growing beyond the core application, and the idea of a mobile companion app makes a lot of sense. The overall approach is very promising, and it’s clear that a lot of thought and passion are going into this project. Community-driven initiatives like this are exactly what make open-source projects so valuable.

That said, I personally would love to see two additional features or options considered:

Manual item creation without AI
It would be great to have an option to add items completely without AI assistance, where the user can manually enter the item name and tags. This would be useful for users who prefer full control over their data, want faster entry, or simply don’t need AI-based recognition for certain items.

Configurable image resizing
An option to resize photos by manually specifying either width or height (with aspect ratio preserved) would be very helpful. This would allow users to better control image size, storage usage, and upload performance, especially on mobile devices.

Overall, this is a really exciting project, and I’m looking forward to seeing how it evolves. Thanks again for your engagement and for contributing to the Homebox ecosystem!

[Duelion]

Thank you very much for the kind words and thoughtful feedback!

Manual item creation without AI

While technically feasible, I'd prefer to keep Homebox Companion focused on its LLM-powered workflow so that it evolves naturally around it, which is its unique future. Maybe once the dust settles I will revisit this idea.

Also, I understand that a official community driven mobile app is already in development and will likely handle manual entry.

Configurable image resizing

Good idea!
Currently, images are automatically resized to max 2048px (longest side) and compressed to JPEG at 85% quality, but only when sent to OpenAI for analysis. So you are right that the original full-quality images are being uploaded to the Homebox server.

Thanks again!

[tankerkiller125]

Just a quick clarification, the mobile app is not official, but also a community driven project. We currently have no plans for an official Homebox app (as it stands at this moment).

[Duelion]

Sorry about that, just edited my post. Thanks for the clarification.

[schale01]

I think I would prefer an app that does not use AI for item creation. I prefer accuracy in this application and the time it takes to proof and correct AI created entries does not Justify its use for me.

[Duelion]

Fair enough.

[crispybegs]

this could work really well, but it does need a manual creations element. I have several raspberry pis in flirc cases, but when i uploaded a photo on the demo site to test it I got this:

It's understandable why the app got it completely wrong, but I would still need to retype every single field so it would be quicker and easier to simply create it manually in the first place.

[Duelion]

Thanks for trying it out and for the feedback, always appreciated.

This is actually a good example of why the AI sometimes doesn't give a satisfactory description. When all it has to work with is a picture of a plain case with no visible branding, it can only describe what it sees. That's intentional, the prompt explicitly tells the model not to guess at brands or models without evidence (you can change this in the settings though, if you don't mind false positives).

Since it works this way by design, I've added a few easy ways to help it along:

1. Add another picture. If there's a label on the bottom of the case, include a picture of that too. Or even better, pop the case open for a photo of the Pi itself. The branding on the chips should be more than enough for the AI, and you might even get bonus fields like the serial number filled in automatically.

2. Give it context beforehand. Before running the analysis, just add something like "Raspberry Pi 4" into the description field. That's enough for the LLM to generate a much better result.

3. Give it context after the fact. If you've already run the analysis, on the Review page you can also correct it and re-run that single item instead of the whole batch. This works escentially like option 2.

There is a learning curve, but after a while you get a feel for when to nudge the LLM along, and when not. For example, I have some plastic picks for prying open electronics, but at first glance anyone would assume they're guitar picks, so I knew I needed to give it a little more context.