Pinned
Cua
782 posts
Cua
@trycua
- Cua reposted
CuaがLinux対応！ AEOもめちゃくちゃ強いし、 Cuaが ComputerUseの覇権をとるぞ。 私が投資家だったら投資するもん - Cua reposted
Wow. I didn't think it was possible. Congratulations @francedot you are actually a wizard. Linux now has Computer Use! - Cua reposted
this is big for people running Hermes Agent on Linux Cua Driver lets your agent drive real desktop apps in the background, through MCP or CLI, while your own desktop stays usable. your agent is now able to use its hands. 
- Replying to @trycua7/ A lot of Linux work is trapped behind desktop apps internal GTK and Qt tools, Electron clients, back-office portals. You can install on Debian 12, Ubuntu 22.04, Rocky 9, and Fedora 41 - code is open-source for anyone to contribute other distros support and report issues8/ Linux Cua Driver is available today. Use it from Claude Code, Codex, or your own agent through MCP/CLI. Repo: github.com/trycua/cua Blog: cua.ai/blog/inside-li…
- Cua reposted
@trycua raising the bar for computer agents. They really said "let's find the hardest possible program to use". Enter PCB CAD software1/ Today we're launching Cua-Bench with @SnorkelAI: a benchmark for computer-use agents on professional software, open for any model to run. The benchmark covers 25 expert-authored KiCad tasks, and the best frontier model we tested cleared only 6 of them.
GIF - Cua reposted
Codex CUA is promising, but I’m finding it unreliable for real workflows right now. Would love to see OpenAI look more seriously at @trycua: open-source CUA infra, agent-ready sandboxes, cross-OS support, and a more practical runtime layer. cc @AriX @JamesZmSun @ajambrosino - Cua reposted
AI #Agents still struggle with #CUA, especially, using complex specialized apps like KiCad. Cua-Bench's EDA tasks developed by @SnorkelAI experts in collaboration with the @trycua team to test agents for realistic engineering tasks that require GUI interactions.1/ Today we're launching Cua-Bench with @SnorkelAI: a benchmark for computer-use agents on professional software, open for any model to run. The benchmark covers 25 expert-authored KiCad tasks, and the best frontier model we tested cleared only 6 of them.GIF - Cua reposted
Computer-use agents are getting strong on daily tasks with Fable 5 reaches 85% on OSWorld Verified. But how well do they work on specialized professional tools? We evaluated Claude on 25 expert-authored KiCad tasks using @trycua’s computer-use framework. Results: • 4/25 full
- Cua reposted
New research: we partnered with @francedot and @ddupont808 at @trycua to stress-test a frontier computer-use agent on real electrical engineering tasks. 25 expert-authored KiCad tasks. 4 passed. 0 build-from-scratch tasks succeeded. The failure modes are concrete and they point1/ Today we're launching Cua-Bench with @SnorkelAI: a benchmark for computer-use agents on professional software, open for any model to run. The benchmark covers 25 expert-authored KiCad tasks, and the best frontier model we tested cleared only 6 of them.GIF
