The Physical Agent Harness: Why Project Mirage’s 'Dune' Keypad Matters
The AI hardware conversation has spent the last year obsessed with wearables—smart pins and pendants trying to replace your smartphone.
But for developers, the real operational bottleneck isn't checking the weather on the go.
It is the physical friction of constant context-switching at the desk: jumping between IDEs, GitHub, terminals, and back-to-back meetings.
This week, Project Mirage—a startup founded by former Ultrahuman VP of Hardware Apoorv Shankar—launched a piece of hardware that approaches this problem differently.
Instead of an ambient wearable, they built Dune: a highly specific, context-aware desk tool that acts as a physical extension of your agent harness.
What is Dune?
Dune is a minimalist, three-button keypad for Mac.
It is tiny (40mm wide), built from CNC‑machined anodized aluminum, and plugs directly and flush into the side of your laptop via USB-C, requiring no battery or external cables.
Unlike a standard macro pad where you hardcode rigid keyboard shortcuts, Dune is context-aware.
It actively detects which application is running in the foreground and automatically updates what its three keys do in real time, without manual profile switching.
If you are in VS Code, the keys might map to Explain, Comment, and Compile.
If you switch to GitHub, they instantly become Approve PR, Request Changes, and Close.
1. The Death of Context Switching
According to the team's Product Hunt launch, approving a standard pull request takes an average of 4-6 clicks.
Multiply that across a full day of code reviews and jumping between tools, and the friction becomes expensive.
Dune removes the navigation layer.
Because it syncs with your calendar and tracks your foreground state, it guarantees that the physical keys under your hand are mapped to the most relevant actions.
If a meeting starts, you don't need to hunt for the Zoom link.
One click joins the call.
A dedicated mic toggle physically unmutes you and instantly snaps the meeting window to the front while multitasking.
2. Agent Triggers on the Desk
The most significant engineering shift here is how Dune handles AI integration.
Dune allows you to map custom AI agents to physical buttons.
If you have an email assistant, a calendar manager, or a code-review agent built in Claude or Openclaw, you can trigger those workflows directly from your desk without breaking your current window context.
We are used to thinking of an "agent harness" as pure software—the sandboxes, the linters, the API boundaries.
Dune extends the harness into the physical world. It gives agentic workflows a dedicated hardware trigger.
🕵️♂️ Reality Check
Whenever a background application constantly monitors your active foreground windows, reads calendar data, and has the authority to execute scripts or approve GitHub PRs, permissions are critical.
Security teams need to treat these physical-digital bridges as attack surfaces.
To make Dune work seamlessly, the daemon requires deep macOS Accessibility and Automation permissions.
If a physical button can approve a deployment or trigger an LLM workflow, least-privilege API scoping for those connected agents is essential.
You do not want a hijacked context-reader gaining broad write access to your repositories.
🎯 The Takeaway
The AI hardware market is finally maturing.
We are moving away from novelty consumer companions and toward highly specific, context-aware tools that reduce operational friction.
Dune proves that the next interface shift isn't just about chatting with models.
It is about bridging the gap between local physical action and complex agentic workflows.
