I had a brokenish keyboard for like 2 months. it would double press keys, but LLMs ain't care. So I ain't care. But now I realize I should've cared. Typing speed is up.
Everytime you see someone complaining about the claude max plan, just imagine anthropic paying you $5000 every month. Because thats what the api pricing is. Did the math yesterday. You can grab statistics from openclaw. $200 is 5 mins of Opus in openclaw.
Yesterday I published winuse, Open-sourced LinUse 🐧 — computer-use for Linux. AI agents drive GNOME/Wayland natively via evdev virtual devices. No Docker, no Xvfb, no fake desktops. If you show up as a touchscreen, Mutter just lets you in 🤷
https://t.co/iZJUrWWMTw
🪟 Open-sourced WinUse — computer-use for Windows. No Mac Mini needed.
Control any Windows box from Claude, Codex, OpenClaw, or whatever agent you're running. Screenshots, mouse, keyboard — just REST.
https://t.co/WYJfi3DleF (https://t.co/WYJfi3DleF)
@msuiche They're both very usable, the issue is, kimi k2.5 is getting hammered so hard, a bunch of openrouter providers are dying. I'm about to try the kimi k2.5 directly from moonshot myself.
Built a gradio with a proper dockerfile and proper cuda layers to run qwen3-tts. Build it yourself or grab the docker off of github https://t.co/BYJBJBPfE6
Lots of the latest and greatest models require the latest vllm, just bumped to vllm 0.14.0 (3 days old) and you can now set VLLM_NIGHTLY true in env to get nightly https://t.co/l56M2ab4Oc