jaredrhod/barehands

What it solves

It transforms a standard webcam into a hand-tracked spatial interface, allowing users to manipulate digital content—such as notes, images, and 3D models—directly on their screen without the need for headsets, controllers, or gloves.

How it works

The system uses Google MediaPipe for hand tracking and three.js for 3D rendering, both loaded via CDNs. It creates a virtual "board" where files from a local folder (a "vault") are represented as floating glass cards.

Users interact with these cards using specific hand gestures (pinching, dragging, flicking, and clapping). The project is designed as a "body" that can be connected to an AI "brain" (like Claude or a local LLM) via simple protocols: the AI can update the interface's state (e.g., changing the visual ring to indicate it is thinking) or trigger specific actions on the board (e.g., presenting a specific note or exploding a 3D model) by writing to small state files or using provided shell scripts.

Who it’s for

It is designed for builders and AI enthusiasts who want a tactile, spatial way to interact with their AI assistants and their personal knowledge bases (like Obsidian vaults).

Highlights

• No specialized hardware: Works with any webcam and Chrome browser.
• AI-Ready: Provides a set of protocols and scripts that allow an AI agent to control the board and monitor its state.
• Spatial Gestures: Includes complex interactions like "the claw" to pull distant objects and two-finger drags to expand 3D exploded views.
• Knowledge Integration: Directly renders Markdown files from local folders as interactive cards.
• OBS Integration: Supports a transparent mirror mode for high-quality streaming and recording with alpha channel compositing.