What AI services are you selfhosting? Or, have tested and passed on

I use OLlama & Open-WebUI, OLlama on my gaming rig and Open-WebUI as a frontend on my server.

It's been a really powerful combo!

Would you please talk more about it. I forgot about Open-webui, but intending to start playing with. Honestly, what do you actually do with it?
- I have the same setup, but its not very usable as my graphics card has 6gb ram. I want one with 20 or 24, as the 6b models are pain and the tiny ones don’t give me much.
  
  Ollama was pretty easy to set up on windows, and its eqsy to download and test the models ollama has available
- I have Linkwarden pointed at my ollama deployment, so it auto tags links that I archive which is nice.
  
  I've seen other people send images captured on their security cameras on frigate to ollama to get it to describe the image
  
  There's a bunch of other use cases I've thought of for coding projects, but haven't started on any of them yet
- I have this exact same setup. Open Web UI has more features than I've been able to use such as functions and pipelines.
  
  I use it to share my LLMs across my network. It has really good user management so I can set up a user for my wife or brother in law and give them general use LLM while my dad and I can take advantage of Coding-tuned models.
  
  The code formatting and code execution functions are great. It's overall a great UI.
  
  Ive used LLMs to rewrite code, help format PowerPoint slides, summarize my notes from work, create D&D characters, plan lessons, etc
- Sex chats. For other uses, just simple searches are better 99% of the time. And for the 1%, something like the Kagis FastGPT helps to find the correct keywords.

None currently. Wish I could afford a GPU to play with some stuff.

Well, let me know your suggestions if you wish. I took the plunge and am willing to test on your behalf, assuming I can.
Yeah. I have a mini PC with an AMD GPU. Even if I were to buy a big GPU I couldn't use it. That frustrates me, because I'd love to play around with some models locally. I refuse to use anything hosted by other people.
- Your M.2 port can probably fit an M.2 to PCIe adapter and you can use a GPU with that - ollama supports AMD GPUs just fine nowadays (well, as well as it can, rocm is still very hit or miss)
- Most LLM projects support Vulkan if you have enough VRAM

I messed around with home assistant and the ollama integration. I have passed on it and just use the default one with voice commands I set up. I couldn't really get ollama to do or say anything useful. Like I asked it what's a good time to run on a treadmill for beginners and it told me it's not a doctor.

Like I asked it what's a good time to run on a treadmill for beginners and it told me it's not a doctor.

Kirkland brand meseeks energy.
- Hey now Kirkland brand is respectable, usually premium brands repackaged. Such as how Costco vodka was secretly ("secretly") Grey Goose
There are some experimental models made specifically for use with Home Assistant, for example home-llm.

Even though they are tiny 1-3B I've found them to work much better than even 14B general purpose models. Obviously they suck for general purpose questions just by their size alone.

That being said they're still LLMs. I like to keep the "prefer handling commands locally" option turned on and only use the LLM as a fallback.
Haha, that is hilarious. Sounds like it gave you some snark. afaik you have to clarify by asking again when it says such things. "I'm not asking for medical advice, but..."

I have Immich that has AI searching for my photos. Pretty useful for finding stuff actually

Once I changed the default model, immich search became amazing. I want to show it off to people but alas, way too many NSFW pics in my library. I would create a second "clean" version to show off to people but I've been too lazy.
- Can't you tag the NSFW to filter it out?

I run ollama and auto1111 on my desktop when it's powers on. Using open-webui in my homelab always on, and also connected to openrouter. This way I can always use openwebui with openrouter models and it's pretty cheap per query and a little more private that using a big tech chatbot. And if I want local, I turn on the desktop and have local lamma and stab diff.

I also get bugger all benefit out of it., it's a cute toy.

How do you like auto1111 as I've never head of it
- The project is a bit out of date for newer models, Though Older ones work great.
  
  I recommand ComfyUi if you want fine grained control over the generation and you like to tinker.
  
  Swarm / Reforge / Invoke if you want neat, up to date UI.

LMStudio is pretty much the standard. I think it's opensource except for the UI. Even if you don't end up using it long-term, it's great for getting used to a lot of the models.

Otherwise there's OpenWebUI that I would imagine would work as a docker compose, as I think there's ARM images for OWU and ollama

Well they are fully closed source except for the open source project they are a wrapper on. The open source part is llama.cpp
- Fair enough, but it's damn handy and simple to use. And I don't know how to do speculative decoding with ollama, which massively speeds up the models for me.

I’ve an old gaming PC with a decent GPU laying around and I’ve thought of doing that (currently use it for linux gaming and GPU related tasks like photo editing etc) However ,I’m currently stuck using LLMs on demand locally with ollama. Energy costs of having it powered on all time for on demand queries seems a bit overkill to me…

Have to agree on that. Certainly only makes sense to have up when you are using it.
I put my Plex media server to work doing Ollama - it has a GPU for transcoding that's not awful for simple LLMs.
- That sounds like a great way of leveraging existing infrastructure! I host Plex together with other services in a server with intel transcoding capable CPU. I’m quite sure I would get much better performance with the GPU machine, might end up following this path!

I was able to run a distilled version of DeepSeek on Linux. I ran it inside a PODMAN container with ROCM support (I have an AMD GPU). It wasn't super fast but for a locally deployed and self hosted option the performance was okay. Apart from that I have deployed Fooocus for image generation in a similar manner. Currently, I am working on deploying Stable Diffusion with either ComfyUI or Automatic1111 inside a PODMAN container with ROCM support.

Didn't know about these image generation tools, besides Stable Diffusion. Thanks!

Try running an AI Horde worker, it's a really great service!

Not sure I know what that is. As in Hoarder?
- It's a cluster of workers where everyone can generate images/text using workers connected to the service.
  
  So if you ran a worker, people could generate stuff using your PC. For that you would gain kudos, which in turn you can use to generate stuff on other people's computers.
  
  Basically you do two things: help common people without access to powerful machines and use your capacity when you have time to use the kudos whenever you want, even on the road where you can't turn on your PC if you fancy so.

If Immich counts for its search system, then there's that.

Otherwise I've tried some various things and found them lacking in functionality, and would require leaving my PC on all the time to use.

What else did you try and what was lacking?

I spun up ollama and paperless-gpt to add ai ocr sidecar to paperless-ngx. It's okay. It can read handwritten stuff okayish, which is better than tesseract (doesnt read hand writing at all), so I throw handwritten stuff to it, but the difference on typed text is marginal in my single day I spent testing 3 different models on a few different typed receipts.

Which specific models did you try and how would you rank each is usability?
- I tried minicpm-v, granite3.2-vision, and mistral.
  
  Granite didn't work with paperless-gpt at all. Mistral worked sometimes but also just kept running sometimes and didn't finish within a reasonable time (15 minutes for 2 pages). minicpm-v finishes every time, but i just looked at some of the results and seems as though it's not even worth keeping it running either. I suppose maybe the first one I tried that gave me a good impression was a fluke.
  
  To be fair, I'm a noob at local ai, and I also don't have a good gpu (gtx1650). So these failures could all be self induced. I like the idea of ai powered ocr so I'll probably try again in the future...

Can anyone suggest a model for light coding? I'm on a 3070 mobile.

Claude is the standard that all others are judged by. But it's not cheap.

Gemini is pretty good, and Qwen-coder isn't bad. I'd suggest you watch a few vids on GosuCoder's YT channel to see what works for you, he reviews a pile of them and it's quite up to date.

And if you use VScode, I highly recommend the Roocode extension. Gosucoder also goes into revising the roocode prompt to reduce costs for Claude. Another extension is Cline.
Qwen coder or the new gemma3.

But at this size using privacy respecting api might be both cheaper and lead to better results.

I am curious about trying an application specific AI. Like just for coding, for instance. I assume the memory requirements would be much lower.

afaik Ollama would fit that bill, but perhaps others can chime in. You could probably run it on your local computer with a small model based on CPU alone.
- I haven’t sunk much time at it, but I’m not aware of any training data focusing on code only. There’s nothing preventing me from running with general purpose data, but I imagine I’d have a snappier response with a smaller, focused dataset, without losing accuracy.