Ask HN: Are you running local LLMs? What are your key use cases?
2025 feels like a breakout year for local models. Open‑weight releases are getting genuinely useful: from Google’s Gemma to recent *gpt‑oss* drops, the gap with frontier commercial models keeps narrowing for many day‑to‑day tasks.
Yet outside of this community, local LLMs still don’t seem mainstream. My hunch: *great UX and durable apps are still thin on the ground.*
If you are using local models, I’d love to learn from your setup and workflows. Please be specific so others can calibrate:
Model(s) & size: exact name/version, and quantization (e.g., Q4_K_M).
Runtime/tooling: e.g., Ollama, LM studio, etc.
Hardware: CPU/GPU details (VRAM/RAM), OS. If laptop/edge/home servers, mention that.
Workflows where local wins: privacy/offline, data security, coding, huge amount extraction, RAG over your files, agents/tools, screen capture processing—what’s actually sticking for you?
Pain points: quality on complex reasoning, context management, tool reliability, long‑form coherence, energy/thermals, memory, Windows/Mac/Linux quirks.
Favorite app today: the one you actually open daily (and why).
Wishlist: the app you wish existed.
Gotchas/tips: config flags, quant choices, prompt patterns, or evaluation snippets that made a real difference.
If you’re not using local models yet, what’s the blocker—setup friction, quality, missing integrations, battery/thermals, or just “cloud is easier”? Links are welcome, but what helps most is concrete numbers and anecdotes from real use.
A simple reply template (optional):
``` Model(s): Runtime/tooling: Hardware: Use cases that stick: Pain points: Favorite app: Wishlist: ```
Also curious how people think about privacy and security in practice. Thanks!
LLMs in general are going to make C/C++ development viable again for the masses. A big reason everything went to web frameworks was memory safety - even big frameworks like Electron and React Native are memory safe overall. Writing low level code by hand was a migraine experience, even for experienced devs. AI changes this relationship entirely, even if you just use it as a pair programmer. I've had even small 8b models correctly call out memory unsafe code and suggest fixes. Larger models are even better.
Local LLMs? A new renaissance. All that power without having to pinky swear with a cloud provider that they won't just take your generated code and use it for themselves.
Expect to see some awesome Windows and Mac apps being developed in the coming months and years. 100% on device, memory safe, and with a thin resource footprint. The 1990s/2000s are coming back.
Thanks-this is genuinely encouraging; I'd assumed AI help was strongest on front-end work(web apps/SwiftUI), so this is my first concrete example of an LLM catching memory‑unsafe C/C++-could you share your toolchain (CLI/IDE integration) and model details (name/quant/runtime), and what "awesome" 100% on‑device Windows/Mac apps you most want to see?
I almost exclusively self host the models I use.
Currently I am using llama.cpp for an interactive repl chat. I was previously using Alpaca (a GTK GUI), but was annoyed with how slow it was and some random crashes. I am transitioning some of this to self hosted in the cloud for things that can't run on my laptop.
I am looking to get away from my current interface, and write my own. Mostly for experience of deeply integrating agents into a program. If anyone knows a good library for interacting with a local model that doesn't involve standing up a webserver I am interested :)
My daily driver is gemma3n. Its been a nice balance between speed and performance without spinning up my laptop fans.
I am super interested in local models, partially because there is no friction from managed services, but also because I think as small models become more viable we will see an explosion of apps incorporating them.
Gemma3n as a daily driver sounds nice—4b or 8b? and rough tokens/sec on your laptop? And have you A/B‑tested code generation quality across local models (e.g., Gemma3n vs others)?
I am using the smaller one, specifically the e2b-it flavor.
I get ~20-30 tok/sec. It's fast enough that its not frustrating, but if it were faster you could more easily skim as it generates.
I haven't done any serious testing. My process is typically learning about new models on HN or elsewhere, and trying to give them a real shake. I have some goto code generation prompts that I try on all of them. None succeed but they are getting close. I also do a lot of just feeling it out. The more I can use solutions unedited the better it feels.
Restrictive privacy laws.
Also log into your Claude/OpenAI dashboard and read the logs. Now they log every damn thing that goes through the API and keep it there for a minimum of 30 days without any option to delete (unless you're enterprise). No anonymization or anything. Just raw audit logs.
Thanks for raising the privacy angle. Do you have a source or plan details for the 30‑day retention and the lack of deletion options (non‑enterprise)? It would help to know account tier and where that policy is documented.
Beyond policy, how are you actually using local LLMs—what tasks do you run locally vs. in the cloud—and which scenarios feel most privacy‑sensitive to you (e.g., proprietary code, contracts, health notes)?
Tried running a local LLM and it felt like adopting a pet dragon. Fun at first, but then it keeps eating all my GPU and still refuses to clean up its own context window.
Haha, a cute pet dragon. Two knobs that helped me tame VRAM: KV‑cache quant/eviction and sliding‑window attention (if your runtime supports them). What model/runtime and context are you running when it tips over? Are you using Ollama?
qwen3-coder for programming help as I'm not very good at css nor flask. sometimes I test other models just for fun, but only for coding or sysadmin.
Python coding is practically the only usecase for local for me.
Cloud llm are able to run 1 trillion parameters and have all of python knowledge in a transparent rag that's 100gbit or faster. Of course they'll be the bestest on the block.
But when the new GPT coding benchmarks only barely behind grok 4 or gpt5 with high reasoning.
>Model(s) & size: exact name/version, and quantization (e.g., Q4_K_M).
My most reliable setup is Devstral + openhands. unsloth Q6_K_XL, 85,000 context, flash attention, kcache and vcache quant at Q8.
Second most reliable. GPT-OSS-20B + opencode. Default MXFP4, I can only load up 31,000 context or it fails?(still plenty but hoping this bug gets fixed), you cant use flash attention or kv or v quantization or it becomes dumb as rocks. This harmony stuff is annoying.
Still preliminary, just got working today, but testing is really good. Qwen3-30b-a3b-thinking-2507 + roo code or qwencode, 80,000 context, unsloth q4_k_xl, flash attention, kcache and vcache quant at Q8.
>Runtime/tooling: e.g., Ollama, LM studio, etc.
LM studio. I need vulkan for my setup. rocm is just a pain in the ass. They need to support way more linux distros.
24gb vram.
Super useful config dump—thanks. Do you have wall‑clock numbers for prefill/gen tokens/sec and power draw on the 24GB card for those three setups? Also curious where quality starts to degrade vs. context length in your tests.
[dead]