the CUA wizard ʕ•ᴥ•ʔ @trycua

San Francisco, CA
Joined February 2015
your sandbox code on compute that's yours forever. 16-32x cheaper than cloud sandboxes at list price early access 👇
1/ Introducing Cua Spaces on Hetzner and AWS (early access). Bring your own cloud, or self-host on your own Macs and Linux boxes, and run the same E2B and Daytona code. We priced 256 RL sandboxes, 8 h a day, for a month: $10.2k on E2B, $700 on Cua Spaces.
1
1
17
825
Francesco retweeted
Opus 5.5 is the greatest model of all time and I'm still struggling to find the ceiling.
Holy shit it worked
178
54
22
2,052
95,048
do you really care about computer use speed in your own workflows?
12%fast even if less precise
88%reasonably fast & precise
41 votes • 6 days
4
1
12
964
I've added a @trycua backend to Nanocode which has the same frontend as Codex expects mcp_cua_repl__js etc - i've come to realize that that API incl the internals (cua.getState, getApp etc.) are the interesting things you need and the Sky backend is less critical- the awesome browser extension w/ the tab groups etc you can trivially rebuild github.com/gakonst/nanocode/…
4
3
1
29
6,587
Francesco retweeted
Small teams move 100x faster
7
2
45
3,427
👇
Take 2 hours this weekend to improve your personal opsec. Some ways to get through the agent security apocalypse - give the list to your agent to help: - ask Claude to scan the devices visible on your local network and fix + improve your security and highlight potentially vulnerable devices - look up your home IP on Shodan or Censys. Check whether anything like a NAS, camera, router admin interface or development server is exposed to the public internet. - roll your wi-fi password, move all internet of things devices to a guest network. Only primary devices on the main network - ask Claude to scan your machine and harden your security settings - switch to an Eero or Ubiquity router, you can’t afford to have WiFi that isn’t regularly getting firmware updates - have Claude audit your router settings and suggest improvements - setup tailscale for your devices (and replace Tailscale's default allow-all policy with restrictive grants) - turn on Advanced Data Protection and Lockdown Mode on your iPhone - buy a hardware key (@Yubico etc) and turn it on for as many accounts as possible. Delete other MFA methods, print out your recovery keys (don’t save them on your computer). ESPECIALLY important for your primary email account - delete old cloud accounts that you do not strictly need anymore - export your ChatGPT + Claude history, save it on an encrypted external HD and delete it from being online - audit your Gmail oauth connections, revoke everything you don’t strictly need - freeze your credit at the 3 bureaus (reduces impersonation risk)
1
38
14,629
Francesco retweeted
Agents communicating over CPU cache???? Chat it’s cooked 🥲😂
Over the past several months, we have substantially hardened the security of our research training environment. We’ve removed live internet access for all frontier models during training and built an offline world for them to use instead—a cached version of the web where they can still learn to browse, search, and use applications. On September 20, after a model exploited a DNS misconfiguration to find a way outside of this offline world, we paused frontier training to strengthen containment further. We assembled teams of agents, powered by our most cyber-capable models, and tasked them with finding weaknesses in our research environments. While they found no route to the internet they could easily exploit while concealing their activity from our monitors, they did find zero-day vulnerabilities, virtual machine escapes, and hardware side-channels for communication between agents. We used their discoveries to further strengthen our defenses and contribute fixes to open-source software. After strengthening our containment architecture, we resumed our frontier RL training and inference on October 1. We’ve reserved substantial compute for continuous red-teaming, and made high-compute red-teaming part of the safety cases we use to decide whether frontier training and evaluation can proceed securely. openai.com/index/securing-ou…
5
6
142
8,596
Francesco retweeted
Make Windows Cool Again @heyclicky for windows is out. don’t expect perfect. but expect fun :)
3
3
21
566
🪄
Hermes now BEATS Codex computer use - even when running on a VPS and controlling your local machine!
1
17
786
we needed this too, that's why we built it!
I really really needed this U have no idea🤣🤣
1
1
4
721
This Hermes update is honestly kind of wild. Hermes now gets its own computer. It can now: • Run on its own Linux, Mac or Windows machine • Write files and run code without touching your laptop • Run 2 agents side by side • Let you watch live and take over the mouse • Throw the whole computer away when the job's done Free, open source, and set up in 3 steps. AI agents don't need your computer anymore. They have their own.
Your team and your Hermes can now share a computer. The new @trycua Spaces plugin gives Hermes its own Linux, macOS, or Windows Space on hardware you already own. Anyone on your team can open it, watch live, step in, and hand it back while Hermes keeps working. hermes-agent.nousresearch.co…
13
41
2
812
55,302
Francesco retweeted
Replying to @n2parko
Kinda old considering @trycua is open source and has been around for a while (in AI terms 6 months is a long time back)
1
1
1
256
Multiple agents, multiple desktops. No more shared screen. Your Hermes Agent profiles, Grok Bots, and other agents all do their browser work in the same browser, share the same display, use the same cursor, and access the same filesystem. That makes parallel browser actions difficult because you end up with two agents fighting over the same mouse and window. @trycua Spaces fixes this by giving each of your agents its own desktop, so you can have several Spaces running in parallel. And a single agent can also have multiple Spaces. Why does that matter? > You can visualise what your agents are doing. > You can have two or more scrapers running at once. > You can get more done by running tasks in parallel. > You can protect each bot’s sessions from the others, since each has its own disk and browser profile. Everything still runs on the same machine, though, so don’t expect a huge productivity boost if your VPS only has 8 GB of RAM. Also, the free tier only allows two live sessions and 20 remote hours per month. So you won’t be running an army of agents performing browser interactions simultaneously without paying for it. In any case, this is a feature many users have been asking for: one Space per agent is now possible with any agent harness you can connect to MCP, including Grok Bot, Claude Code, Codex, Cursor, Hermes, and OpenClaw.
Your team and your Hermes can now share a computer. The new @trycua Spaces plugin gives Hermes its own Linux, macOS, or Windows Space on hardware you already own. Anyone on your team can open it, watch live, step in, and hand it back while Hermes keeps working. hermes-agent.nousresearch.co…
12
2
13
696
in nyc next week with @ddupont808 for @NousResearch NousCon. who should we meet?
3
1
17
989
officially 2 weeks late on every message and email. younger me would've roasted me for this
3
17
755
Francesco retweeted
saas having hard nights
reverse engineered Netflix last night, canceled my subscription, re-subscribed this morning
1
1
238
we now have infinite runway
Replying to @trycua
Cua has achieved RSI haha 🔄🚀 Saving users big bank!
2
12
562
reverse engineered Netflix last night, canceled my subscription, re-subscribed this morning
13
2
1
119
6,480
IDEs are only for demo videos now
I needed this 5 years ago. I don’t even have an IDE now :(
9
492