Parallax

@tomosman @openclaw @NousResearch mac minis are underrated for this. we've been running multi-node setups on apple silicon with parallax and the performance-per-dollar is hard to beat. nice to see more people building this way.

Infinitely bullish on a stack of MacMinis or Studios at home running @openclaw or @NousResearch Hermes. Run local models and soon you will have AGI at home. Lots of other epic stuff too but feels downstream of being able to do this.

@cyb3rops the "local" label is doing a lot of heavy lifting for some of these apps. if your data still round-trips to someone else's server, it's not really local. with parallax, your inference actually stays on your machines. your devices, your models, even offline.

Can anyone explain this to me? First Claude Workspace, then Perplexity, now Manus - they keep using words like “my”, “personal”, and “local” in a way that suggests local information isn’t being sent to a remote LLM or RAG system for evaluation. But if no local LLM is actually running, then almost nothing except maybe config stays local. The reasoning still happens remotely. Right? Also - does anyone really think this belongs on a corporate workstation?

some parallax dev lunch break fun: - a macbook pro, a mac mini, some cables - zero internet, zero cost - openclaw running on parallax no subs. no token burn. nothing leaves the desk. just local agents vibing.

@AlexFinn yes. that's why parallax exists.

If you have your OpenClaw working 24/7 using frontier models like Opus, you're easily burning $300 a day. That's $100,000 a year. I have 3 Mac Studios and a DGX Spark running 4 high end local models (Nemotron 3, Qwen 3.5, Kimi K2.5, MiniMax2.5). They're chugging 24/7/365. I spent a third of that yearly cost to buy these computers I'll be able to use them for years for free On top of that they're completely private, secure, and personalized. Not a single prompt goes to a cloud server that can be read by an employee or used to train another model I hope this makes it painfully obvious why local is the future for AI agents. And why America needs to enter the local AI race.

messari's new report on echo-2 highlights how parallax powers the rollout plane. consumer RTX 5090s served as distributed rollout actors via parallax, feeding a centralized learner cluster. and we got 33-36% lower hardware costs with no quality loss. this is parallax doing what it was built for. turning consumer hardware into production AI infrastructure.

@farukomerekinci @perplexity_ai this is the right framing. but owning the stack means owning the inference too. openclaw + cloud api = local agent, cloud brain. openclaw + parallax = local agent, local brain. that's the difference between "your data stays on your machine" as marketing vs as reality.

Here's the difference with OpenClaw: Perplexity: Their AI, their servers, your data through their pipeline. One model. One product. Take it or leave it. OpenClaw: Open source. Runs any model, Claude, Grok, Kimi, whatever you want. Your data never leaves your machine. You build the agents, you set the rules, you own the stack. What's now on the table, with a $1B company validating the category: -AI that checks your email before you wake up -Agents monitoring your business 24/7 -Cron jobs running strategies while you're offline -Your entire workflow automated, on hardware you own The difference between Perplexity's version and what you can build yourself isn't features. It's control. Perplexity = Shopify. OpenClaw = owning the server.

Announcing Personal Computer. Personal Computer is an always on, local merge with Perplexity Computer that works for you 24/7. It's personal, secure, and works across your files, apps, and sessions through a continuously running Mac mini.

perplexity just announced always-on ai running on a mac mini. the category is real. the question is whether your always-on ai should phone home to someone else's servers or run entirely on your own hardware. we built parallax for the second option.

mac studios + dgx spark + openclaw is super solid local stack. the inference speed and context window issues you flagged are exactly what we're focused on with parallax. automatic sharding across mixed hardware and paged attention on mac make a noticeable difference on both. happy to help if you want to test it alongside your current setup!

After some time of using local AI cluster (Bob), here is my honest take on the good, the bad and overall use case. About a year ago I started playing with local AI models because of the work we do at BottleCap AI. I realised how amazing it actually is to own my own stack and my own data. At first, we used local models mainly because of security reasons as we do lots of AI efficiency research and new product concepts based on that. After OpenClaw was released, something changed for me. I started using local models much more, until they replaced cloud models for most of my deep-thinking tasks beyond work. Eventually, I canceled all my AI cloud subscriptions just to see if I could actually run fully on my local cluster. Hardware: • 2x Mac Studio with M3 Ultra and 512GB unified memory, 32-core CPU • 1x NVIDIA DGX Spark, added recently for prefills and, hopefully soon, faster inference • 10GB LAN Switch for connecting Spark and Mac Studio’s Current models: this is changing pretty frequently 1) “Bob OG”: • Main brain for reasoning and daily tasks • Qwen3.5-397B • Roughly 40-60 tokens/sec (depends on load & task) 2) “Bob Researcher”: • Long term researching • Qwen3.5-27B-Claude-4.6-Opus-Distilled-MLX-4bit: Very experimental 3) “Bob App Developer": • Coding apps and debuging • MiniMax M2.5 Software stack: • OpenClaw: All-local assistant layer • LM Studio: Running models • Exo Labs: Connecting multiple machines into one cluster and testing whether inference improves Where my local stack still lacks: • Deep tasks with big models still take more time to reply than cloud models. • Context window is limitation in the models I use. I’m usually around a 200k token window per session, but compacting works well, so I rarely need to start a new session. • It also seems that OpenClaw in its default state is not handling work with memory very efficiently while filling the context window fairly quickly by default. It was necessary for me to finetune this manually including semantic search and temporal decay which are in default switched off. • Reasoning is good but not at the cloud models level. Also coding is good for the majority of tasks but not top tier. My best use cases right now (March 2026): Best for iterative work where privacy matters and where model needs to be available all the time. • Private or sensitive data: I would be careful as a company to share private or direct customer information with third party cloud systems in general. Clearly also connecting OpenClaw to cloud models is not solving privacy situation. • Cloud limits & Efficiency: If I push cloud subscriptions hard, I hit consumer limits surprisingly fast. It’s also much easier to spot inefficiencies locally. When the context starts bloating, the system slows down fast, so issues like memory inefficiency become obvious much earlier. In the cloud, replies often feel just as fast, but you end up paying much more or hitting usage limits without really knowing why. Was it worth the money? For me, yes. But I’m aware I live in a niche bubble for my particular use case. For most people it is still early. For businesses and people who want to spend the money and effort make this work it is good solution today. My verdict: For my personal use case, local is now the default. Cloud is the exception. Are local models as good as the best cloud models? No. Are they good enough to be my default for most tasks? Yes.

a 9b model now matches a 120b on reasoning benchmarks. models are shrinking fast. the hardware you already own is becoming more powerful every release cycle. this is the trend parallax was built for. connect your machines, run what fits, and scale when you need to.

ancient romans built public libraries changed the way how knowledge is accessed. today, open source AI is doing the same job. open doors, open intelligence. 🏛️

We also tested the messier setups. Using Parallax, we trained Qwen3-8B on distributed RTX 5090s. 36% cheaper than centralized A100s, same scores, zero divergence. We even trained a 0.6B agent to beat LLMs at No-Limit Texas Hold'em. Reliable results with unreliable compute.

They crashed. They fell. They exploded on the pad. Then they got back up. Faster, wiser, stronger. Breakthroughs don't come from one perfect run, they come from the freedom to fail 100 times. Introducing Echo-2, distributed RL that boosts AI research throughput by 10x.

The wait is over. Open your Clawbox now! The easiest way to own your🦞agent. Zero code. Just download, click, and go. Grab yours: commonstack.ai/clawdbot?ref=x

spoiler alert 🦞 we’re preparing a fully private version of clawbox. air-gapped ready. run it completely offline. absolutely zero risks tolerated with sovereign AI. 🛡️

Want an @openclaw🦞agent in just a few clicks? Meet Clawbox🦞, powered by Commonstack. -No CLI. No VPS. Just download & run. -Permission controls for peace of mind. -Switch between 30+ LLMs instantly. Free credits + 20% off for all! 🦞Join whitelist: commonstack.ai/clawdbot

proud of kimi. kimi k2.5 better when sovereign.

heard you all buying mac minis now? should have been grinding harder back then😋

@gradientintern @Gradient_HQ let's popping champagne when antropic open source their 1st model 👀

run locally and never get revoked 🦞 full sovereignty, save dollars and protect privacy. ./ @tryParallax by @Gradient_HQ

what are you building with your mac mini today? hint: save your $5 for vps and have some privacy with parallax.

congrats! we need our own infra for AI talents.