ori-platform/ori-runtime

Ori-runtime is currently in a pre-seed/prototype stage, evidenced by its 14-day age, 4 stars, and lack of forks. It addresses a very real and significant problem: the 'safety gap' in agentic AI when applied to physical environments. While standard LLM agents can hallucinate harmlessly in a chat box, a hallucinating IoT agent could cause physical damage (e.g., turning on a stove). The project's 'graduated authority model' is a novel attempt to bridge this. However, the project lacks a moat. It sits in a space where it must compete with the massive gravity of Home Assistant (which is moving rapidly toward local 'Year of the Voice' and AI features) and the cloud-to-edge offerings of AWS IoT Greengrass and Azure IoT Edge. Furthermore, Apple and Google are both well-positioned to dominate the 'local agentic home' via their existing hardware ecosystems (HomePod, Nest) and their push into on-device AI (Apple Intelligence, Gemini Nano). The defensibility is low because the code is currently a thin wrapper around existing local LLM executors and GPIO controls; the true value—the safety framework—has yet to prove it can foster an ecosystem or provide a 'data gravity' that prevents users from switching to more established IoT platforms.