4/
We are actively seeking contributors and collaborators as we continue to build training-free tools for spatial understanding. Issues, PRs and DMs are open.
We built the kernel on VGGT-SLAM (MIT SPARK Lab), Sam3, DinoV3, DepthAnything. Thank you so much for the amazing work!
1/ Open Reality is the first open source agentic harness for spatial understanding.
Scan a room with your phone. Your AI assistant can then measure it, plan paths through it, and export it as robot-training data.
https://t.co/1UeonCjPah
3/ What it does today:
• video in, 3D scene out
• measure distances and angles (relative units until you calibrate with one real length, then metres)
• path planning through the scanned space
• LeRobot / GR00T dataset exports, Isaac Sim scenes
• scene agents that survey and label
• 41 MCP tools, offline simulator, self-hostable
2/ It will empower Claude to understand and modify engineering work that requires 3D vision: robotics sim training, CAD, game development.
Install:
claude mcp add openreality -- npx -y openreality-mcp serve
codex mcp add openreality -- npx -y openreality-mcp serve