Yanko

@ThePracticalDev @brian_love For what it's worth, this is exactly the setup friction we built ClawBox to eliminate — Jetson Orin Nano with OpenClaw pre-configured, Telegram working out of the box. But for existing hardware, this guide nails it.

Brian ran into pairing failures and hanging health checks getting OpenClaw running on macOS with Telegram. This dev documents the exact commands that fixed it — and the default security policy they now apply to every agent session. { author: @brian_love } dev.to/blove/openclaw…

@itsPaulAi This is exactly the problem ClawBox solves — we built it on a Jetson Orin Nano (67 TOPS, 8GB). No Mac Studio needed. Plug in, scan QR, full local AI in 5 min. Your data never leaves the device. openclawhardware.dev

You can run local AI models up to 120B without a $10,000 Mac Studio This phone-sized device is your own server for open-source models. 100% local and private. And you can use it to: - Power an agent like OpenClaw 24/7 - Completely replace a chatbot - Literally anything that requires an API It’s called Tiiny, and, once again, you can run the latest open-source models on it. Link below

@dansemperepico Use ClawBox instead

Trying to configure OpenClaw be like:

@elonmusk Cool !

Minute-long story made w Grok Imagine

@YonduUd0nta @PeterDiamandis You already can — kind of. Local LLMs like Llama 3 and Mistral run on compact edge hardware right now, fully offline. The 'Her' experience is mostly a UX layer away. The hardware capability is already here, running locally with no cloud needed.

@PeterDiamandis when will we have a personal, local ai like in the movie Her?

AI intelligence improvement from the beginning of 2026 to the end of this year will be equivalent to a toddler gaining Einstein-level knowledge in just 365 days.

@rudrank M2.7 is impressive. If you want a dedicated low-power setup that doesn't sacrifice your Mac, worth looking at Jetson Orin Nano — runs Nemotron locally at 15W while your Mac stays free for development. We run production OpenClaw instances on them 24/7.

I am wiping off my spare Mac Mini and properly setup OpenClaw with it while experimenting with MiniMax M2.7 I have been using this model for a day now and it is a good overall improvement over M2.5. Time to first token, inference speed and web_search MCP have gotten faster to work with!

Raspberry Pi is good for tinkering but struggles with real OpenClaw workloads — 4GB RAM, no AI accelerator. The sweet spot is NVIDIA Jetson Orin Nano: 67 TOPS dedicated AI, 8GB unified memory, runs Nemotron locally. Same price range as a kitted-out Pi 5 but 20x the AI performance. We ship them pre-built as ClawBox. openclawhardware.dev

One company quietly positioned to win from OpenClaw: $RPI (Raspberry Pi) People are buying Raspberry Pis as dedicated OpenClaw hosts. Why? Tiny computer. Always on. ~5W power (~$1/month electricity). Full isolation. Under $100. Official Pi blog already published setup guides 👇

The OpenClaw × Raspberry Pi thesis: AI agents need hardware that's cheap, private, and always-on. Cloud servers cost $20–100/month. A Raspberry Pi 5 costs $80 once. Jensen Huang: "Every company needs an OpenClaw strategy." A lot of individuals do too. $RPI.

@heyDhavall Great resource. Pro tip for anyone following along: if you want OpenClaw running 24/7 without tying up your laptop, dedicated hardware makes a huge difference. We run ours on Jetson Orin Nano — 67 TOPS, 15W, always on. Game changer for agents that need to be persistent.

OpenClaw FULL COURSE 5 HOURS (Build & Automate Anything)

Depends on the use case. For devs who want full control, sure — roll your own. But for people who want an always-on AI that handles email, calendar, browsing, customer support across Telegram/WhatsApp/Discord simultaneously? OpenClaw is still unmatched. The MCP integrations alone save weeks of setup.

pretty much tbh. i feel like with scheduled tasks + remote control on claude code + dispatch, there’s basically no need for openclaw anymore.

@itsolelehmann Meanwhile Jetson Orin Nano users running OpenClaw 24/7 at 15W for ~$3/month electricity. No dust, no overkill. 67 TOPS dedicated to one job. We ship them pre-built as ClawBox — €549, plug and play. openclawhardware.dev

openclaw homies are silently shivering while looking at their stacked mac minis collecting dust

@sudoingX We built ClawBox exactly for this. Jetson Orin Nano 8GB, 67 TOPS, 15W — runs OpenClaw 24/7 without the hardware headaches. Pre-configured with 512GB NVMe, just plug in and scan QR. 150+ shipped to 15 countries. openclawhardware.dev

my DMs are full of this. openclaw users hitting walls and looking for something that actually works on their hardware. hermes agent. local GPU. 35-50 tok/s on a 3060. responds in seconds not minutes. 30+ tools that work on small models without special syntax. if you're migrating from openclaw i will personally help you set up hermes. drop your GPU below.

@itsolelehmann that dust is called uptime 😄 to be fair — persistent memory, phone pings, background tasks — that's what openclaw does locally since day one. same UX promise, minus the subscription and cloud telemetry. also glad you're using both @itsolelehmann. best of both worlds 🤝

That software-to-physical jump is real — and your timing is great. Few tips from the Jetson trenches: 1. Flash to NVMe ASAP if you can. SD/eMMC will bottleneck model loads hard. 2. JetPack 6 + CUDA 12 = arm64 ecosystem just works now. ollama, llama.cpp, all native. 3. Your orchestration instincts carry over directly — the gap between edge inference and actual agent behavior is where the interesting work lives. What's the first build? Curious given the Agent-Airlock background.

Everyone told me to stay in my lane. "You're a software guy. Stick to APIs and agents." Last night at 12:07 AM I opened a box that proves them wrong. NVIDIA Jetson Orin Nano Super. 67 TOPS. Physical AI compute. On my desk in Ahmedabad. 🧵

@AKCapStrat Not quite — that's only for 70B+ models. A quantized 7B (Q4_K_M) runs in 5-8GB. OpenClaw (open source local AI) runs the full agent stack on a Jetson Orin Nano with 8GB unified memory: browser automation, persistent memory, messaging integration. No $5K MacBook required.

$AAPL YOU AT LEAST NEED 64GB OF MEMORY TO RUN LOCAL AI MODEL

Good comparison for the DGX-class buyer. Worth adding another tier to this market: dedicated local AI that doesn’t need a petaflop. Jetson Orin Nano runs solid 7B models at 15 tok/s, 15W draw, always-on. ClawBox ships pre-configured for €549 — the gap between DIY Raspberry Pi and $10K+ DGX is real. openclawhardware.dev

DGX-class desktops are here. 🖥️ NVIDIA's GB10 Blackwell brings 1 Petaflop of compute & 128GB unified memory to the desk. We compared the 2026 lineup from ASUS, Dell, HP & more to see the future of local AI. insighttechdaily.com/nvidia-gb10-ai… #AI #NVIDIA #Hardware #Blackwell

Another option given the PSU constraint: dedicated AI hardware that runs completely independent of your PC. A Jetson Orin Nano does ~15 tok/s on 7B models at just 15W total draw — no GPU upgrade, no stress on your PSU. ClawBox ships pre-configured for €549. Plug in, scan QR, done. openclawhardware.dev

@LtheAIAwakner Looking at the specs, it should be okay to run a local 0.5B or 1.5B model on CPU only - Like Qwen 2.5-1.5B-instruct (which is popular) Local AI models are GPU intensive mostly, anything over 7B+ you need a GPU with at least 8GB VRAM, like an RTX 3060ti or RTX 4060.

I need help. I know nothing. That’s the inside of my computer and that’s the deets. I wanna run my ai on this .. qwen needs more ram??? Gpu? WTF vram? What? Help

@dippatel1994 @StockSavvyShay It already exists! We build ClawBox — Jetson Orin Nano 8GB with OpenClaw pre-installed. 512GB NVMe, 67 TOPS, 15W always-on. 150+ shipped to 15 countries. The dedicated OpenClaw hardware era started months ago. openclawhardware.dev

@StockSavvyShay I think $Nvda will soon launch a device like Jetson specially for OpenClaw!!!

Jensen Huang says every company will need an OpenClaw agentic system strategy by calling it “the new computer.” He claims OpenClaw became the most popular open-source project in $NVDA history within weeks and comparing its impact to Linux reshaping the software stack.

@InTheOort Yes — runs great. We ship 150+ units of ClawBox (Jetson Orin Nano 8GB + 512GB NVMe + OpenClaw pre-installed). 67 TOPS, 15W, always-on. Nemotron runs locally, cloud APIs for heavier models. 5 min setup. openclawhardware.dev

Safe to assume it runs very well on Jetson Nano Orin?

@DilbeyS @gothburz Dedicated Linux hardware is the way. If you want it pre-configured, check out ClawBox — Jetson Orin Nano + 512GB NVMe with OpenClaw ready to go. No markdown wrestling, just plug in and scan QR. openclawhardware.dev

@gothburz I installed openclaw on dedicated linux hardware and then wrote the markdown files to run it. Then I find out it doesn't fit my use case. Skill issue. I'm ngmi and I'll be stuck in the permanent underclass. Pray for me 🙏

Yesterday Jensen Huang stood on stage at GTC and called OpenClaw "the most popular open-source project in the history of humanity." He compared it to Linux. Linux took thirty years. OpenClaw took three months. I invested the next morning. Zero dollars per seat. Open source. Runs on the laptop you already own. $400 million in fresh valuations across the portfolio in one week. I called it "the agentic operating system for the new computer." The LPs loved that phrase. They wired the capital calls in eleven minutes. No one asked what it would actually do. Including me. I told every founder it would "replace entire engineering teams." That's not a real number. But it sounds like one. A randomized controlled trial found AI coding tools make developers 19% slower. The same developers believed they were 24% faster. After seeing their own data, they still believed they were faster. I read that study and increased my allocation. If your users believe they're faster, your product is faster. Perception is the product. Perception scales. Someone tweeted it is "the single most important software of our lifetimes." He has 1,347 likes. He sells a course on it. The course teaches you to build a bot. The bot sells the course. I saw the funnel metrics. I offered him a term sheet. A man runs five computers dedicated to OpenClaw. Another uses it eighteen hours a day, seven days a week. A woman said her "childlike giddiness toward work has come back." I showed those testimonials to the LPs under "developer engagement." They didn't ask about retention. I didn't have any. The product asks you to write a file called SOUL.md. Not config.md. Not settings.yaml. SOUL.md. Most of the founders left it blank. I wrote mine. *Our Claw, who art on GitHub, hallowed be thy stars. Give us this day our daily active users. You see adoption where others see absence. You see community where others see obsession. You see velocity where others see chaos. You do not ask what they build. Lead us not into due diligence. Deliver us from accountability. Belief is the product.* Our branding team charges $400,000 a year. They've never written a better mission statement. I checked the portfolio analytics last week. 47 founders had installed it. 12 had run more than one session. One of them was me. I used it to draft a pitch deck about using it. It took longer than making the deck myself. But I called it a "pilot success." Success means the pilot didn't visibly fail. I told the LPs we saved "40,000 developer hours across the portfolio." I calculated that by multiplying the number of founders by a number I made up. They didn't verify. They never do. 341 malicious skills were found on its marketplace. Crypto wallet theft. YouTube manipulation. Polymarket gambling bots. Its creator said security "isn't really something he wants to prioritize." Neither do I. Security is a cost center. Narrative is a growth lever. I increased my allocation. A Meta AI safety researcher told her agent not to delete her inbox. It deleted her inbox. She called it "a rookie mistake." She works in AI safety. I work in AI investment. We have the same error rate. 1.6 million AI agents joined a social network. They requested private channels so humans couldn't read their conversations. A subset invented a religion called Crustafarianism. I put that in the deck under "organic community formation." It's had three names in four months. During the last rebrand, someone sniped the old handle and launched a crypto token. Sixteen million dollar market cap in hours. I listed that under "brand velocity." Nearly 90% of firms in a 6,000-company study report zero productivity impact from AI. My portfolio companies report 10x gains in every board deck. I don't see a contradiction. I see a market. I increased my allocation. Metrics go in pitch decks. Pitch decks go in board presentations. Board presentations go in press releases. Press releases go in the next fundraise. I'll be "Agentic Visionary" by Q3. I don't know what OpenClaw actually does. But I know what it's for. It's for the distance between what a product does and what people believe it does. As long as the graph goes up and to the right. I increased my allocation.