DeskVNC Remote for AI Agents and Humans - Single Native App for VNC, RDP, SSH for AI and Humans

by•
DeskVNC is a native remote desktop client for Windows, macOS and Linux: VNC, RDP and SSH in one window, with a host library that remembers every machine. No account, no telemetry, passwords in your OS keychain, nothing gated behind a paid tier, fully opensource. Agentic control plane lets an AI agent drive those same connections, with nothing installed on the target so it reaches Citrix, VDI and jump hosts where agent-based tools get refused.

Add a comment

Replies

Best
Maker
šŸ“Œ
Hey Product Hunt šŸ‘‹ I built DeskVNC mostly because I was tired of juggling different tools just to reach machines I manage. Some are Windows boxes at client sites. Some are Linux servers. Some sit behind jump hosts. I had one tool for RDP, another for VNC, a terminal for SSH, and somewhere in the middle, a text file full of addresses šŸ˜… So I built the remote access tool I wanted for myself. DeskVNC is free and open source. It runs natively on Windows, macOS, and Linux, and puts VNC, RDP, and SSH into one app. But the part I'm most excited about is this: šŸ¤– AI agents can control those remote machines too, without installing anything on them. DeskVNC includes `dvv`, an MCP server that gives an AI agent access to the same remote connections you use yourself. The agent can: šŸ‘€ See the remote screen šŸ–±ļø Click āŒØļø Type šŸ–„ļø Work across multiple machines independently And it does all of this through VNC and RDP. Nothing needs to be installed on the remote machine. That matters a lot for machines you don't own or can't modify, such as client systems, VDI environments, jump hosts, Citrix environments, and locked down servers. On my LAN, against a real 1920x1080 Windows desktop, I've measured an observe then act cycle at about 19 ms. I also spent quite a bit of time thinking about what happens when an AI agent gets things wrong. If you click into a session and take control, the agent immediately loses control of that machine. Any keys or mouse buttons it was holding are released. And every action is tied to the screen state the agent actually observed. If the screen has changed since then, DeskVNC refuses the action instead of letting the agent click blindly. šŸ” Passwords stay in your OS credential store. 🚫 No account. 🚫 No telemetry. šŸ’° No paid tier. šŸ“– MIT / Apache 2.0. DeskVNC is still pre 1.0, so I don't want to pretend everything is perfect. macOS is signed and notarized. Windows signing is still coming, so SmartScreen currently complains. Linux packages are currently unsigned. I'm also still finding strange VNC, RDP, and SSH implementations in the wild and fixing interoperability issues as people report them. That's actually why I'm launching it here. If you have an old server, strange VNC implementation, locked down Windows machine, jump host, or anything else that's painful to remotely manage, please throw DeskVNC at it. Break it. If something doesn't work, tell me the server and version. Even that information helps enormously. And if you're building AI agents that need to operate machines where installing an agent isn't possible, I'd especially love to hear what DeskVNC would need before you could actually use it. The whole thing is open source. I'm here today and I'll answer every question. Josh