Show and tell: VibeFuse — a free Windows canvas that runs Qwen Code (plus Claude Code, Codex, Gemini CLI, Cursor) as live PTY widgets #11754

Founder post — I built this tool. VibeFuse is a free Windows desktop harness that runs Qwen Code as a live, draggable widget on a canvas, alongside Claude Code, Codex CLI, Gemini CLI, and Cursor. One window, all your coding agents, each in its own named PTY session.

Why Qwen Code users might care:

• Named sessions, cloneable. You can create, rename, clone, mute, or close a Qwen Code session without killing the underlying PTY. I use clone to run two Qwen Code sessions against the same repo when comparing prompts, or one Qwen + one Claude Code session on different halves of a task.
• Canvas instead of terminal soup. Widgets remember their position and session tabs persist the layout. Terminal, Browser, YouTube and Files-explorer widgets share the same canvas, so agent output sits next to what you're actually working on.
• Voice stays local. Whisper STT + Piper TTS run on-device (in C:\vibefuse), with an optional wake word + hold-to-talk dock. No voice traffic routed to a cloud service.
• MCP panel — Google Workspace/Drive one-click plus custom stdio servers, attached to the same widget sessions.

There's also a marketplace for agent skills and widgets (Stripe Connect; sellers keep 80%, skills install to C:\vibefuse\skills). That's how the free harness gets funded — the harness itself has no subscription and doesn't resell model access. Agent CLIs remain yours; VibeFuse is the session/canvas/voice/marketplace layer.

Honest current state: v0.1.24 public preview, Windows 10/11, and the installer is unsigned for now (I can share the installer SHA-256 if you want to verify before running it).

Download: https://fuseintelligence.org/products/vibefuse — docs: https://fuseintelligence.org/docs/vibefuse — marketplace: https://fuseintelligence.org/marketplace

Three things I'd genuinely like Qwen Code users' feedback on:

1. Is a Linux/macOS port worth prioritizing, or should Windows features come first?
2. Which Qwen Code workflow gaps hurt most day-to-day — checkpoint/restore, non-interactive mode, subagent model switching, something else?
3. Would a Qwen-specific section in the skill marketplace make sense? What would you want in it first?

I'll be around to answer everything.