OpenCastor

The product is an open source runtime that connects AI models to robot hardware and translates natural language into bounded, signed motion plans, solving the integration problem of pairing different AI providers with different robots without custom glue code. It is for developers and business teams building robotics applications, including engineering teams working with hobby to industrial hardware. It is positioned as a universal adapter built on the RCAN specification that normalizes AI providers to a standard interface via YAML configuration, with local models, human-approved limits, and offline-verifiable receipts.

Key features

• Natural language to bounded motion plans
• Single YAML config for AI and hardware
• Support for 12+ AI providers
• Hardware-accelerated vision via Hailo-8
• Human-signed plans before execution
• Human-approved budget, window, speed cap
• Spoken stop bypasses model stack
• Deadman stop after ~1s without commands
• Ed25519 signed receipts for every action
• Offline receipt verification via script
• Pairing via QR code with castor up