Hey all! Just wanted to start a good thread here on what AI tools you are running in your home lab. Here is my shortlist:
- Ollama
- OpenWebUI
- AnythingLLM
- WhispherX
- Stable Diffusion
- ComfyUI
Extending the conversations from my last video to this forum thread:
I'm running some in local Docker Desktop and also Proxmox LXC container with GPU passed through. What about you?
Believe it or not, still rocking the GTX 1060 6GB! Check out my blog on this project here: Still Worth It? Running AI Workloads on a GTX 1060 in 2025.
Have a RTX 4060 in my workstation currently that I run the --gpus all parameter with.
Great Video and super interesting topic. Been digging into this for some time myself as a hobbyist and homelab noob. I discovered another tool that may be of interest to you. Fabric by Daniel Meissler and his team. Operates on text, documents. web pages, YouTube videos, etc. and there are a couple of plugins for obsidian that allow using it within that environment. My favorite one is MeshAI. Fabric's uniqueness comes from its "patterns"-highly detailed community created prompts. Can be customized or can create your own. You can even just download their patterns for use in other GPT applications. These patterns can be chained together to create workflows as well. Pretty neat tool.
@kevinmorris, I totally agree in the super interesting topic for sure. Hey are you referring to the fabric.so tool? Wondering though if there is another tool by the same name? I did try this one out as an evernote replacement, and really liked it. I think this one is definitely one to watch. I still thought it could use some polish overall, but the AI capabilities are great compared to something like Evernote. I need to take a look at MeshAI, haven't had my hands on that one as of yet. Are you using these tools mainly for documentation and AI capabilities with your notes? I am curious what use cases everyone is finding to be helpful.
Thanks Kevin!
@brandon-lee, yeah, the same or similar name, but it is a different project. It's funny, there are like 4 different projects all using the name fabric. LOL. I should have added the link originally to clarify. The one I am referring to is at this link.
it continues to grow and expand in number/types of patterns and functionality.
@brandon-lee, my primary use case is for summarizing and analyzing my notes in obsidian as well as boiling down and summarizing YT videos. Then when I find interesting ones, I will go watch the whole thing. I also create workflows where I can take a technical video and summarize pulling out key learning points to throw in my notes for reference later. I still take some notes by hand as well, but this kind of augments this. However, there are all kinds of patterns that I have not even delved into that can be used for all sorts of different use cases. once you see what these patterns consist of and what others have done, it brings ideas for others.
@kevinmorris Again, great insights here. Are your workflows happening all within Fabric? Or are you using something like n8n to orchestrate the steps? Not sure on the capabilities of the Fabric project as of yet that you have mentioned here, so this may be answered in the docs, just need to look through. I'm a huge fan of n8n so far, but definitely looking at other projects and tools out there and this one seems super interesting.
@brandon-lee it's within fabric, ut have also dabbled with similar flows using the patterns from fabric in make. You have me curious about n8n after this. Video. Will be looking into this as well. Doing it in fabric is just stringing together patterns via pipes in the commands. In msh Ai plug in obsidian you can also do this with maybe a bit more flexibility. Basically use prompt to extract transcript, then can run another to summarize and another to extract wisdom. Then run last one to grade the summarized results compared to the transcript. This step is more playing than anything. I was playing in make to do this as well. Run transcript through multiple LLMs and then compare results to see which one yielded the best results via a grade. All messing around mostly to learn the tools.
Another one that is interesting is Msty.ai. I have not played with this one for a while, and when I checked to see tis current status before mentioning it here, it seems it has grown tremendously since I used it. It is now available on linux via appimage and debian installer. Going to have to put this on my list of things to dig into again. It was just on windows when I was dabbling before, and I was in the process of setting up a linux focused homelab. When I was checking all of these different apps out at that same time, it compared to Anything LLM in its approach, but I liked Msty better. Can't recall my reasoning though as it has been about 6-12 months.
@kevinmorris I have heard of Misty.ai, need to take a deeper look at it though as part of my next round of things to try out.
Currently running Claude CLI on my 3 proxmox nodes. It leverages my local ollama instance that uses 2/4 gpus from a Tesla M10. It also leverages Gemini and OpenAI. I can talk directly to it from my pve shell or through my personal Jarvis website I built. This system can enact changes and audits directly on cluster and services. I also have it set to run an audit and create improvement recommendations. It’s quite nice to have something fill the gaps in knowledge and quickly bring lofty ideas into fruition. My wife appreciates it because I’m not in my personal office as long just doing research.
