What does the agent actually do on screen?
With Uremot, you can watch it work live in your browser. If your agent is testing a desktop app, you see the same screen as it clicks through the workflow. The agent can hear desktop audio and use the keyboard and mouse - all of this in sandboxes.
We recorded this 15-second demo across three workspaces on one Linux desktop. The opening streamed at 1080p and about 60 fps with both desks on the same machine.
Don't limit your imagination - Give your agents hands, eyes and ears - then watch.
Dedicated CPU cores, separate kernel, better isolation.
Sounds great, but how much overhead is too much?
We've been measuring this at @Uremot_com.
Curious where you'd draw the line.
☁️ Existing customers can request Cloud early access via Console → Get Help.
Uremot cloud ☁️ on the way 🚀
We ran one Small agent desk as a VM and as a container on one server.
Encode latency at p50, desk spread over the machine: VM +0.7 ms idle, +1.9 ms moving.
On one whole core, as we'll run it: both slower, and the VM missed our 2 ms line when moving.
@lahfir@Uremot_com sandboxes work on any spare hardware you have or cloud VPS you use. Interact with your Agents through MCP or with uremot CLI. Bring your own harness. You watch your desks from browser.
If you're running VMs on shared cloud infrastructure, CPU scheduling is worth a closer look.
While benchmarking microVMs at @Uremot_com, we found something unexpected.
The interesting part? Linux core scheduling and SMT.
Benchmarks and raw data:
https://t.co/sbKVmNreoh
@OpenAI Using the new gpt-6.2-sol and gpt-6.2-astra on @Uremot_com and the token usage is just crushing all the expectations. A single 100$ plan covers me for 200$ opus-5.5 subscription even with reset. Thanks OpenAI!!!
@ctatedev Checkout Uremot: https://t.co/NiOFIgeqvS
You can self-host your own agents, on your machines, on your models or harnesses in containerized sandbox environments.
Your agent needs a computer. It won't have to be yours.
Coming: Uremot Cloud, machines we run for your agents. Each desk in its own VM, on whole CPU cores no other customer's desk runs on. Not available yet, and self-hosting stays.
From a fresh machine to your first agent in one command.
uremot setup connects Claude Code, Codex or OpenCode, and runs each sign-in. Then, hire your first agents.
🤖 Give AI agents hands, eyes & ears.
Introducing Uremot: real desktops for AI agents, running on your machines.
Let agents browse, code, use apps & test their work.
Watch live with sound. Take control anytime.
Bring your own model.
🔗 https://t.co/xN24SOA813
#AIAgents
Hate texting yourself to move text between devices?
I've built DecentPaste. Copy on laptop, paste on phone Works on desktop and mobile
E2E encrypted, local only. No cloud
Open source. Rust, Tauri, libp2p
https://t.co/ismGyuNi11
#rustlang#TauriApps#libp2p#OpenSource#selfhost
🎉 Scala 2.13 is here! Simpler, faster, safer collections is the centerpiece. And: library enhancements, literal types and other language features, a faster compiler, and lots more. 1500+ pull requests from 162 contributors were merged. Let’s celebrate! https://t.co/I8RJLbMCXy
Hey everyone👋 I’m thrilled to announce Bloop v1.3.0 is out! 😻
Let me quickly summarize the key features and improvements that make this release so special for Scala developer tooling👇