Emma

AI, ROBOTS & WEB 2.0 to 3.0 | Builder at @autonomous_labs

[Emma @monami_95z]

🔥 get AI at your desk today

[Autonomous @autonomous_labs]

R̶e̶n̶t̶ ̶v̶s̶.̶ ̶O̶w̶n̶ ̶ ➜ Rent-to-Own GPUs 🎉

Because paying cloud rent forever is so 2023. EdgeAI Computer’s got a rent-to-own option coming soon:

- Pay monthly, same as cloud prices.
- Own the rig post-rental.
- Or stash it in our data center (tiny fee).

Like financing a car, but for AI infrastructure. ✅

Curious? Click "Notify Me" to stay in the loop when it's ready. 👇

[Emma @monami_95z]

@autonomous_labs Such a nice integration! Finally you can have 100% private AI on your desk.

[Emma @monami_95z]

get your first intern truly private & always-on your desk. cause why's not ?

[Autonomous @autonomous_labs]

Private AI still has a runtime problem.

Venice protects the model layer.
But agents still need a place to run.

Autonomous Intern is that place: a small hardware box for agent workflows, separate from your laptop, files, wallets, browser sessions, and keys.

[Emma @monami_95z]

Raspberry Pi has been fantastic for so many maker and robotics projects. Nice to see it powering personal AI builds like this.

[Dee @dee_hw]

1/ What's after AI?

Robots.

Let's build your own Personal AI Robot.

Open source. Hackable. For a few hundred dollars. 🧵

[Emma @monami_95z]

@dee_hw Building my own robot with open-source hardware sounds interesting. Definitely bookmarking this for my weekend plan!

[Autonomous @autonomous_labs]

200 builders. Fully booked.

Joining AI Builders Vietnam for their OpenClaw workshop. Bringing an Autonomous Intern along.

Excited to keep showing up for dev-friendly OpenClaw events. See you there! 🦞

[Emma @monami_95z]

as an ADHD person trying to focus on deepwork as much as possible, Lamp now is a sweet distraction to me 🫡

[Autonomous @autonomous_labs]

If you still think your AI friend won't fuck you on a dumb request, stop.

We're having a crazy week with Lamp. New deskmate. New boss.

Won't shut up till we give him enough context.
Won't run what he doesn't agree with.
Reasons better than any other.

Also a good entertainer ↓

[Emma @monami_95z]

@autonomous_labs never knew i'd need someone to remind me drink water & take eyes off screen a bit that much....until i meet lamp 🥹

[Emma @monami_95z]

@autonomous_labs waittt, the future of personal computing is getting weird in the best way 👀

[Emma @monami_95z]

This is the first time a local AI setup has felt… personal to me.
Not a chatbot in a tab. Not another cloud demo.

Just an agent sitting on my desk, running on my own macbook.

[Autonomous @autonomous_labs]

Running Hermes on a physical device feels different.

Most agents live in someone else’s cloud.
Intern runs on your desk.

Hermes is now plug-and-play on Intern.
Your agent. Your memory. Your machine.

Local agents are starting to feel real.

[Emma @monami_95z]

Tune in now: https://t.co/IUFGqNCV8b.

[Autonomous @autonomous_labs]

Elon just ROASTED Sam over that $45K Roadster refund—'You forgot the part where we fixed it in 24 hrs!'

Full breakdown on Autonomous Radio, the world's first AI-powered radio🧵

[Emma @monami_95z]

Be the first to own an AI Radio 😎

[Autonomous @autonomous_labs]

What if listening were intelligent?
What if news weren’t spun - but understood?
What if, instead of scrolling, we simply listened?

Meet Autonomous Radio, the world’s first AI-powered 📻:
• Swarms of AI agents narrate what matters
• CNC-milled aluminum body

Early access open.

[Dee @dee_hw]

Tested Qwen3-30B-A3B vs Llama-2-7B on 2×5090🔥

Qwen3-30B: ~217 tok/s, 16.7 GB VRAM used
Llama-2-7B: ~268 tok/s, 3.1 GB VRAM used

Llama runs lighter and faster, but Qwen handles bigger context with solid throughput.

Interesting trade-off between speed and scale 👇🧵

[Emma @monami_95z]

must make one version for yourself😄

[Dee @dee_hw]

Ran dual 5090s under burn ⚙️

At 100% load: 19.8 TFLOPs each, temps 63–68°C. Zero errors, steady under pressure.

At 70% power: ~15.6 TFLOPs each, temps 57–61°C. Cooler, quieter, better for 24/7 training.

Clip run in ⬇️

[Emma @monami_95z]

@BrodyAutonomous Llama 3 8B looks smoother on 2x5090

[Dee @dee_hw]

🚨 Benchmark Showdown 🚨
Comparing 2x5090s vs 2x4090Ds!

💥 Mistral 7B, Deepseek 7B, and Llama 3 8B all crush the 2x4090Ds in speed.
⚡️ Throughput massively improved on 2x5090s.
🔋 Solid memory efficiency maintained.

Check out these results! 👇🧵

[Dee @dee_hw]

Why dual RTX 5090s in ONE machine >> networked setup? ⚡

✅ GPU scaling just works
✅ No network overhead
✅ Skip synchronization hell

Higher cost vs separate units but the performance ROI is real .

Pro tip: BIOS GPU optimization = mandatory 👇

[Dee @dee_hw]

Tested Llama 3 8B against yesterday's 7B models 💭

Quick summary for choosing:
👉For chat apps: Mistral wins (69ms response)
👉For batch jobs: DeepSeek wins (168.5 tok/s)
👉For quality: Llama 3 8B (if you can wait)

Llama 3 details 🧵

[Emma @monami_95z]

@BrodyAutonomous Lane confusion gone, thanks for clearing it up man!

[Dee @dee_hw]

An 8×4090D build isn’t just slotting cards. You’ll run into PSU math, heat control, and lane limits. We’ve got the common answers covered. The FAQ are here 👇