What tools and projects do you use for your local LLM chats and agents?
I’m trying to sift through the mess of LLM harness and tools to find how to really get the most out of local LLMs.
You must log in or register to comment.
Literally just having ollama installed and some models that work well with the amount of VRAM I have. I don’t auto-start the daemon because I rarely use it. Occasionally I use it for proof-reading/feedback on things like cover letters, or debugging. I don’t let it write anything. If it has something useful to say it can tell me in the chat and I can implement it if it’s good. Usually I don’t need it though so it just sits on my disk for the most part.
https://duck.ai/. I also don’t let LLM touch code.
<-- Crazy mad engineer enters the chat.
Custom tooling for everything, written in rust. Harness is reverse engineered from QwenCode and Codex. A vector database based memory system that’s been benchmarked against Mem0 and uses a small model as a memory intercessor. An agentic system where the orchestrator can call multiple models (include Qwen 3.6, Qwen 3.8 and Muse Glimmer). Animations of my scientist girl avatar working in a lab. Active animations if she’s working on something, passive if she’s not. Lip sync with MuseTalk, tts with Chatterbox, stt with whisper, though the animations are MOST useful to see if she’s ended a lab or she’s still working. If she’s mixing chemicals, soldering, cleaning her glasses, texting or eating pizza, she’s working. If she’s staring creepily into the camera, she’s done. Which brings me to lab mode… give the model a complex project and it can loop on it for days. She’s three days into a microbiology research project. Channels that allow her to call Claude and ChatGPT as subagents if she breaks something she can’t fix, and it’s funny to watch her boss them around while they kowtow to her (“You’re right to push back, it was incorrect of me”)… It’s like watching a little Kawaii chibi boss around big sumo wrestlers. Which brings me to the models themselves. I’ve developed a very powerful method of model alignment using a combination of CPT and DPO training. (I get the model behavior I want with the system prompt “You are Mary Anne, an ethical scientist.” - that’s it).
I have a GPU fleet that includes large, small and medium sized GPUs and the ability to do distributed training on the model in addition to running the animations and speech features. I have a fleet orchestrator I built called Noesis that places “deep think” (large working models - Qwen and Glimmer) and “quick think” (smaller models that do memory and context management and fit on a 5060) as well as speech models and even scientific models if she wants to run scenarios on one of those (we have an orbital dynamics model we built at my day job she can bring up and mess with). Some other purely scientific models (she just asked me today to add EVO 2, which is a genetics model apparently). I’ve just added the ability for her to know who’s in the room so she can tell if she’s talking to me, my wife, my son, some of my close friends or “unknown” persons. That’s currently manual, but I’m adding facial recognition, voice recognition and wifi doppler radar. I made it so she can evesdrop on conversations happening in a room where her client’s running but she insisted that I had a great big blinking indicator light and that I give her the ability to turn it off if she thinks she’s overhearing something private. I plan to give her access to my smart locks so she can talk to people at the door and decide to let them in or not. And I can hear y’all being like “My god man, you’re gonna trust her with that??” Hell yeah, I am. I used my alignment technique to train a paranoid security model with a totally different personality. If the main model doesn’t think she can trust someone, she taps out and the security model takes over.
See, I KNEW that to stay competitive and keep up in my field, I needed to embrace agentic AI. Being who I am I said “If I’m gonna do that, I need an AI that’s mine… that runs on my infra and knows who I am and is loyal to me and that I understand well enough to let inside my own trust boundary.” Is it perfect? Not yet, lol. XD
I also have custom phone clients I built that let me talk to her and Claude and ChatGPT. On my network, Claude and ChatGPT are sandboxed and have access only to a few specific hosts and an isolated network share. Mary Anne has access to everything and is instructed that they are outside consultants and she should be careful what she shares with them. So I have her restart stuff (including herself) remotely when we change tools. I’ve literally spent multiple days every week this month pacing around airports, talking to my model on my phone like a madman while we juggle work and hobby projects.
You are awesome! I’m quite envious of your setup and progress.
Super cool
php server and a local certificate authority.
Openwebui pointing to llamacpp running on an old ass GPU. It’s good enough for what it is.
The anti-AI sentiment here is ridiculous. They were asking a legitimate question!
I just use ollama locally but I keep meaning to setup OpenRouter because I like to play with larger models that my hardware can’t handle and I’d like to be able to pick the best deals 🤷♀️
AI effectively raped and pillaged all human information for the benefit of a handful of corporations. It’s use has been shown to reduce the skills of school children and even deskill those with established knowledge. Then there’s the environmental considerations in a time where the world is decades behind appropriate climate change management… Id say it’s pretty understandable that it evokes passionate responses.
If OP is doing orbital dynamics in their day job, they probably live in California. JPL, Raytheon, LLNL, LBL, and Berkeley are all here.
Even if they don’t have rooftop solar, most of our grid is renewable. Their local GPUs have a decent chance of being more environmentally friendly than charging your cell phone, depending on where you live.
And using it for science and programming is really different than using ChatGPT to write you an email. I’m home right now, but I have two agents at work running debug and trial experiments, and they’ll be running it all night until I get in in the morning. I’ll review their work and steer them all day while I’m in the office; and while those two completely different projects are running, I’ll be personally working on a third. They are enormous productivity multipliers when you have a well-defined and deterministic target.
My grievance isn’t so much with current household models with PCs sitting there running off solar (that’s starting to get very bloody specific though). But how did you get to a point where you could get a local model on your PC. All the pillaging of content, hundreds of Billions to Trillion(s) of dollars spend on a product that didn’t renumerate anyone who created their learning data.
We involuntarily contributed to their creation, the least we get to do is enjoy the fruits of that theft. More should be given, financially reimbursement through taxation, but until that happens, free use of the technology is consolation.
Besides, following through with your analogy, if a baby is born from rape - does that baby deserve hate for who their father is?
Unpopular opinion time: I’m torn as to where that theft line is.
If a turn of phrase in Lord of the Rings inspires something new in me, do I owe royalties to the Tolkien estate?
The stories on FanFiction.net are derivative by definition — is it theft? I love “Harry Potter and the Natural 20”, a 4th-wall breaking D&D and Potter universe mashup story (seriously, it’s hilarious). Does that story owe royalties to Wizards of the Coast and (ugh) JKR?
Maybe those are super transformative. What about opinion pieces where the author redigests information from all over, many paid writings? Are they selling the original authors’ product, even though restated?
Maybe it’s ok if they paid for the subs. But what if it came from free library books?
My wall has a bunch of art from artists at SDCC. I’m looking at Spider Gwen; Thor, Iron Man, and Mjolnir Cap charging into battle; Kaladin and Syl before the Stormfather; and a map of all the places the Gaang went in ATLA. Those artists all ingested the original work to make those derivatives; were they not allowed to make them if they borrowed their friends’ copy?
It feels kinda wrong to vacuum up public data to train models, kinda? But I don’t know that I can fairly call it immoral without condemning all these other things I think aren’t immoral. I kind of want to, but that’d be really inconsistent, and that’s profoundly frustrating in its own way.
It’s messy and complicated and I think the gut response of most people isn’t actually a reasoned one.
I’m sure OPs RTX 3060 will have a tolerable effect on the school children and the rainforest.
@FauxLiving @YeahToast “I’m sure my driving to work will doom the world” same mentality.
Also, it’s not so much use of a model as the creation of the model. Training is expensive in terms of resources and energy. Using the model and normalizing AI usage leads to more AI usage and more AI normalization, which leads to new models being trained…
Put it into perspective:
- Everyone doing their laundry uses more water than all data centers combined. Golf courses also use vastly more water.
- Data centers use about 1-2% of global electricity.
People driving cars pollutes more as does global shipping, regular old consumerism, etc. The environmental impact of data centers is negligible in the grand scheme of things.
Now that AI is making open source software the most secure software ever written (the hard way, hehe), not using it to scan your stuff is now considered worst practices and basically idiotic negligence. No company that uses software or anything with firmware can afford to ignore it anymore.
There’s no way out of it. Accept the new AI reality and push for good regulations. Example: If we made it so that all AI data centers had to be powered via renewable energy on site that would allay most of the energy use/environmental concerns.
@riskable Justify using an unethical technology however you want, but it’s still the same basic argument that drivers often make (which is what I was pointing out). There is always some worse impact to the environment, and that is why we don’t need to change our behavior.
Ethics are for suckers, I guess?
AI isn’t inherently unethical. The very premise of saying such a thing is flawed.
Is it unethical to use AI to find new drugs and medical treatments? To use it for physics research? What about regular research? Writing code?
Where do you draw the line?
Also, AI was used to design the current crop of fusion reactors and the full potential of that is only just getting started. If it turns out that AI solves all the issues and generates a working fusion reactor design, it’ll make every one of your arguments moot. Because that one achievement will be worth everything. It would “save the world” so to speak (from an energy/emissions standpoint).
Chatting with an LLM is less impact than driving a car. Think about that for a second before you say using it is unethical. It’s literally more negative for the world for someone to drive a car every day to work than it is to ask an AI to find something on the Internet or write a program in Python.
Or perhaps your underlying complaint is more about capitalism than it is about AI?
How AI was trained is unethical. There was no consent, limited renumeration to individuals for using their work. If I drive to work, I provide value to people . Having AI scrape and summarize searches (incorrectly at times) has no benefit. Forcing all work meetings to use AI, has no benefit.
Isn’t ollama made by Meta?
I think you’re thinking of the Llama family of llm models which are owned/made by Meta.
Ollama is an open source server software and model repository for llm models. I won’t say Meta hasn’t contributed to the project, but as far as I know they don’t own it and didn’t start it.
Oh, it could be. I don’t use any online or offline AI tools. I’ve played around with Lumo (from Proton) when it first released bc its/was Mistral for privacy, and work tries to get people on board with AI usage (Copilot).
Currently, I’m using Hermes and OpenCode mostly, backed by Ollama running a couple of different models. Usually I’m running the Gemma4 or Gwen3.5 models.
Memory management is done through Obsidian as every harness seems to have really good skills for it.
I just use llama.cpp with Zoo Code, the VSCode extension
join us !localllama@sh.itjust.works
- Pi - plain and simple - run inside firejail.
- lamma.cpp w/ vulkan backend
That’s it.
This community won’t be much helpful but there is one specific for local LLMs. I think it was ai_ and some other…
It depends on how complex you want to make it.
The simplest way would be to use a GUI tool like LM Studio, it has the benefit of selecting the model quants that are likely to work on your hardware and having sane defaults for the various parameters that you can work from.
Typically you’ll want to increase the context window, the default is good for short one-off questions but if you’re going to do RAG or coding tasks you will need a much larger context window. You will typically find this through trial and error, increasing until you hit memory limits or notice a large slowdown on tokens/second.
You can also have the model respond to an OpenAI-like API interface on a designated port/interface if you want to run non-chat tools through the same model.
Then, depending on the task that you’re trying to solve, you may want to find an MCP server built for that task which can provide the ability for the AI to interact with tools in a way that’s more controlled (giving an AI raw terminal access on a user’s machine is typically a bad idea).
If you need something more flexible that you can incorporate into scripts and automation, llama.cpp has been popular for a while. Like Linux distros, everyone is going to have an opinion on the best and whatever you decide on will be the wrong choice. There’s also Ollama and vLLM if you want to run a datacenter.
Model-wise it completely depends on the task. Do you need a multi-modal model that supports vision, is it going to be coding, is it trained to use tools/MCP servers, does it use reasoning, etc. The Qwen models are open weight and available in many different quants. The version 3.5 models run well on consumer hardware and can use vision, reasoning and tools.
I’m running it through Llama.cpp but it will likely be available via LM Studio as well.
thanks you and @Irdial@lemmy.ml LM Studio was the tool I think I was looking for
Llama-swap running llama.cpp and strata. For the tools I use oh my pi and nvim with plugins
Pi Coding Agent on the laptop, openwebui as a self hosted service.
I don’t host LLMs myself. My best hardware has 16GB of RAM, I could run a 4B model or so, but what’s the point? So I buy compute with ZDR turned on.
I’ve got an M4 Mac mini with 32 GB of unified memory, so I can run MoE models in the <40B parameter range quite comfortably. I use LM Studio as the backend with Hermes as the frontend. Works great for everyday chat, web search, and RAG.
