Sooo I went through this exact same journey last year when I started hitting the API limits on the web versions. I really wanted that premium feel without the $20/month sub, especially since DeepSeek-V3 and DeepSeek-R1 are so dirt cheap to run. Basically, I spent weeks hopping between different setups because I'm kinda obsessed with low latency and clean UI.

Just sharing my experience: I actually started with Open WebUI like everyone else, but I found it a bit heavy for my older Synology DS920+ NAS. It's awesome for features, but the Docker image is huge. I eventually switched over to LobeChat and honestly, it was a game changer for my technical workflow. It's basically a polished frontend that feels almost exactly like Claude.

Here’s why it clicked for me:
- The mobile-responsive design is literally better than most paid apps.
- It handles LaTeX math rendering perfectly for my physics notes.
- File uploads for RAG are super smooth; I just drag in a PDF and start asking questions.

I mean, you gotta be careful with the database setup if you want persistent history across devices, but if you just want a snappy Docker container for DeepSeek endpoints, it's worth a look. I've also tinkered with LibreChat which is more "ChatGPT-clone" style, but Lobe felt more modern for daily use... just my two cents though! GL with the setup!

Respectfully, I'd consider another option before settling on the big names. Honestly, while Open WebUI is the gold standard for features, it can be a total resource hog if you're trying to keep things lightweight on a budget VPS or an older home lab setup. In my experience over the years, I've found that LobeChat is actually the hidden gem for DeepSeek-V3 and DeepSeek-R1 users.

I mean, I've tried many interfaces, and LobeChat feels sooo much snappier than AnythingLLM. Here is why I think it fits ur workflow better:

1. It's built for speed: The UI is super polished and feels highkey like a premium native app. The mobile-responsive design is actually usable, unlike some other Docker-based UIs that get weird on small screens.
2. DeepSeek optimization: It handles system prompts and LaTeX rendering perfectly out of the box. If you do technical work, the math output is super clean and doesn't break.
3. Budget-friendly RAG: It has built-in file support and knowledge base features without needing a massive vector DB setup.

So yeah, I'd suggest trying the LobeChat Docker Image first. It's literally one command to deploy and the API key integration is super straightforward... definitely the smoothest daily driver I've found lately! gl with the setup!!

Honestly, I've tried a bunch of these and it's been a total rollercoaster. I started with some basic scripts but eventually moved to Open WebUI, and while it's super popular, I actually had some issues with it feeling a bit bloated for my setup. It's reallyyy powerful, but sometimes the RAG felt a bit clunky with DeepSeek R1.

For your situation, I would suggest checking out LobeChat. It's literally the smoothest experience I've found for DeepSeek V3 and R1. The UI is honestly gorgeous—very ChatGPT-like—and it's super mobile-responsive. It handles LaTeX rendering like a champ for technical work, and the Docker deployment is basically one command. Plus, it has a dedicated plugin system if you wanna expand it later. Another solid alternative if you want something ultra-light is Chatbox AI, though it's more of a desktop app than a self-hosted web UI. But yeah, if you want that polished web feel with easy API integration, LobeChat is the way to go imo. gl with the setup!

Saved for later, ty!

Tbh, I've spent way too much time benchmarking these setups. If you want that "professional" feel without the bloat of the more mainstream options, you should definitely look into these two alternatives: 1. LibreChat: This is basically the ultimate power-user choice. It supports DeepSeek via the OpenAI-compatible endpoint and handles system prompts way better than simpler wrappers. The LaTeX rendering is super solid for technical docs, and the file upload/RAG setup is much more robust. It’s a bit more involved to set up with Docker Compose, but it’s worth it for the stability.
2. NextChat: If ur looking for something ultra-lightweight, this is the one. It’s a single-container deployment and the UI is incredibly snappy. It’s perfect for DeepSeek-V3 because the latency is almost zero. It might lack the deep RAG features of the heavier suites, but for pure coding and math, it’s basically perfect. Honestly, I’d go with LibreChat if you need a daily driver for serious technical work. It’s open-source and the customization for different model parameters is exactly what a DIY enthusiast needs.