RIP yaksbeard

A crucial read for ANYONE that uses social media. Send to a friend or loved one as well. wired.com/2017/02/dont-b…

@X_Arcadian What do you have in your settings for it?

@RIPYaksbeard I'm using LM Studio to run Qwen3.5 35B A3B IQ4_XS, its been working pretty well but I've not pushed it hard yet. Just got it up and running.

What are some decent settings (and model) for running hermes using a local model on a 4090?

@KJJKETH @Teknium @sudoingX Yea zero care about honcho... thats why i asked about better memory.. the last thing i want to do is have to plug my shit into 50 different 3rd party services to get functionality out of it.

@RIPYaksbeard @Teknium @sudoingX Honcho is a subscription service. Not a fan of subscriptions.

jensen just compared openclaw slop house to linux and called it the most popular open source project in history. i admire jensen but he has clearly never used openclaw on a small model. if his team had spent one day in my DMs watching people migrate off it to hermes agent because their tool calls kept failing he might have framed things differently. openclaw's founder left for openai. the codebase is 125K+ lines of typescript bloat. the sandbox blocks the tools that actually matter. small models can't use the MEDIA: syntax so your images never arrive. i know because i found that bug, wrote the fix, and got it merged into hermes agent the same day. you don't need a $4,699 DGX Spark or a corporate "openclaw strategy" to run an autonomous agent. you need a half decade old GPU sitting in your drawer and a framework that actually works from 7B to 70B without special syntax. hermes agent. 30+ tools. 11 model specific parsers. runs on a RTX 3060 at 35-50 tok/s. the fix i submitted yesterday is already in production. jensen i respect the vision but the migration is already happening and it's not going in the direction you announced.

@KJJKETH @Teknium @sudoingX Hermes memory is what is keeping me from trying it... give it a better memory system and ill for sure give it a go.

@Teknium @sudoingX OC has better memory. Less framework means more flexibility. Hermes is strong. Just rigid. I have them both on the same directory so they essentially share files.

@sudoingX @StoutimonO Can we get a better memory system for it?

@StoutimonO install it first and tell me where you get stuck. drop screenshots or errors and i'll walk you through it.

@justalexoki Yea give him a sub to a vpn

nephew started pirating movies so sister in law asked me for help to block all those websites but i don't even know any of the big ones anymore. can you guys help me?

@13arm13arm @JWSPEED2 @elonmusk @aelluswamy @Tesla @TeslaAUNZ these people are so out of fucking touch lmao... WHY DIDNT YOU JUST BUY A NEW CAR

@JWSPEED2 @elonmusk @aelluswamy @Tesla @TeslaAUNZ Are you buying me a new car?

7 years ago, I paid for FSD ... I still don't have it @elonmusk @aelluswamy @tesla @TeslaAUNZ Oh, and before you say $2800?! EAP was $8300AUD

@HustleBitch_ Looks like hes wearing a wax mask of his face on his face.

This is the face of a man who just realized he made the biggest mistake of his entire life.

@sudoingX I see you talking about hermes more but im taking a look at it and it seems its ootb memory is kinda garbage? Then in the docs it suggests using some agentic cloud memory? Is there any decent work on better persistent memories?

i work till 4-5am testing models and writing posts so people going local ai don't have to figure it out alone. then you see someone copy paste your entire post word for word with no credit. no RT. claiming it as theirs. pocketing impressions on work they didn't do. i don't mind inspiration. i mind the copy paste with zero appreciation. 4 different accounts today alone. my community has been catching them, writing community notes, and calling them out. that means more to me than the views they stole. the work doesn't stop. they copy posts. i build the next one. but i'd be lying if i said it doesn't sting.

@lilchiva @sudoingX @boyuan_chen Spiritually breakeven

@sudoingX @boyuan_chen I mean, I'm interested and I run local models. But saying that spending a couple of grand on hardware is an instant "break even" is ridiculous.

let me get you started in local AI and bring you to the edge. if you have a GPU or thinking about diving into the local LLM rabbit hole, first thing you do before any setup is join x/LocalLLaMA. this is the community that will help you at every step. post your issue and we will direct you, debug with you, and save you hours of work. once you're in, follow these three: @TheAhmadOsman the oracle. this is where you consume the latest edges in infrastructure and AI. if something dropped you hear it from him first. his content alone will keep you ahead of most. @0xsero one man army when it comes to model compression, novel quantization research, new tools and tricks that make your local setup better. you will learn, experiment, and discover things you didn't know existed. @Teknium maker of Hermes Agent, the agent i use every day from @NousResearch. from Teknium you don't just stay at the frontier, you get your hands on the tools before everyone else. this is where things are headed. if you follow me follow these three and join the community. you will be ahead of most people in this space. if you run into wrong configs, stuck debugging hardware, or can't get a model to load, post there so we can help. get started with local AI now. not only understand the stack but own your cognition. don't pay openai fees on top of giving them your prompts, your research, and your most valuable thinking to be monitored and metered. buy a GPU and build your own token factory.

@l33tdawg @steipete How does it determine what to add to memory and how/when to retrieve?

@steipete @RIPYaksbeard Try this clawhub.ai/l33tdawg/sage-…

There's a lot of cool stuff being built around openclaw. If the stock memory feature isn't great for you, check out the qmd memory plugin! If you are annoyed that your crustacean is forgetful after compaction, give github.com/martian-engine… a try!

@pjain @steipete Its not user error its garbage

Thanks. I will have to try this out. I haven’t gotten great results from using qmd (probably user error here). I’m also burning tokens because for some strange reason, claw won’t use Ollama for embeddinggemma and keeps using OpenAI. Integrating local MLX-audio capabilities for tts and stt have proven to be inconsistent so far but still working on it.

@sudoingX @JoelDeTeves @Teknium You can run local models you control on your own hardware with openclaw.

@JoelDeTeves @Teknium never used anything that starts with "open" or "claw" and never will. i run local models on my hardware i control.

okay the fuss around hermes agent is not just air. this thing has substance. installed it on a single RTX 3090 running Qwen 3.5 27B base (Q4_K_M, 262K context, 29-35 tok/s). fully local. my machine my data. first thing i did was tell it to discover itself. find its own model weights, check its own GPU, read its own server flags, and write its own identity document. it did all of it autonomously. nvidia-smi, process grep, file writes. clean execution. the TUI is genuinely premium. dark theme, ASCII art, color coded tool calls with execution times, real time streaming. you actually enjoy watching it work. 29 tools. 80 skills (that's what it reports on boot). file ops, terminal, browser automation, code execution, cron scheduling, subagent delegation. and it has persistent memory across sessions. setup took 5 minutes. one curl install, setup wizard, point to localhost:8080/v1, done. dropping qwopus for this test btw. distilled models compress reasoning and lose precision on real coding tasks. base model only from here. more experiments coming. octopus invaders (the same game that broke qwopus) will be built using hermes agent next. comparing flow and results against claude code on the same model. if you want to run local AI agents on real hardware this one deserves a serious look.

@samhenrigold The fuck are they buying? Willy wonka inventions?

We’re acquiring Promptfoo. We’re buying Safetysnoot. Proud to announce that the Ragschlorp team is joining us. We own a majority stake in Chunkwad.

@claudiocasalex @sudoingX They do send telemetry but not every prompt. At least not yet.

@sudoingX Do you imply that OpenAI can watch every prompt on OpenClaw even if we use only local agents? How?

buy a GPU. a single 24GB card fits more intelligence than you think in 2026. call APIs wisely. stop giving away your cognition for free. people keep asking about claw and openclaw bots. let me be clear. i will never run anything connected to OpenAI. that includes their arrogant false-confident codex models and their newly acquired openclaw bot. none of it. they are watching every prompt. every idea. every thought you feed them gets augmented into their next product. you are the training data. when i need frontier reasoning i use opus. other than that, everything runs on my hardware.

@hlitoroy @sudoingX I have no clue..

@sudoingX Why do u think opus is safe?

@g_keffa @DecentCloud_org @sudoingX No but a 100$/month sub in 1 year is going to be significantly better than now. That 5090 will have better models but not keeping up with the curve. Especially when they close source qwen

@DecentCloud_org @sudoingX Well are you going to stop using AI after 1 year?

@realmikolson Carrey is well known for fucking with his fans so it won't surprise me if its part of a bit.

Couple possibilities here - he had a butt load of plastic surgery - he completely lost his mind - he completely lost his mind and also had a butt load of plastic surgery - he’s doing a bit and messing with people - he’s had a butt load of plastic surgery, lost his mind, and is doing a bit to mess with people - this is not Jim Carrey, but a clone, because the real Jim Carrey had too much plastic surgery, lost his mind, and was doing too many bits

@RituWithAI @TheAhmadOsman I still find it hallucinates quite a bit tho.. but yes its quite decdnt

GLM-5’s "smart but slow" energy is basically that one brilliant professor who takes ten minutes to find their glasses before dropping a life-changing lecture. Those massive active parameters are a heavy lift, but if accuracy is the priority over speed, the trade-off is often worth the wait.

> GLM-5 is very smart & capable, main drawback: massive active parameters, makes it very slow > MiniMax-M2.5 is small, fast, and smart; my go-to general agent > Kimi K2.5 is still the best opensource multimodal > Qwen 3.5 is like 95% of K2.5 at 1/2 size - just a tad less smart