
Agentic IoT runtime

Agentic IoT runtime
Ori is an open source agentic IoT runtime designed to give physical devices the ability to reason about sensor data and act autonomously. Unlike traditional IoT platforms that primarily collect data, apply thresholds, and trigger alerts for human response, Ori processes sensor events through a tiered reasoning system. This system combines deterministic safety rules with local small language models (SLMs) to enable decision making directly on the device. The runtime operates offline first, with an offline capable safety core, and optional gateway escalation for connectivity dependent tasks. It is built to run on constrained hardware, such as a Raspberry Pi, and is intended for environments with unreliable power, intermittent connectivity, and limited resources. The core of Ori is its Physical Actuation Trust framework, which defines what an AI agent is permitted to do in the physical world. This framework uses a graduated authority model that specifies action consequence levels and required human oversight. The runtime pairs every reasoning decision with an action decision, ensuring that autonomous behavior is always governed by a trust boundary. Key technical features include a hardware circuit breaker in all adapters to isolate failing buses, a model file (Qwen2.5-0.5B Q4) of approximately 500MB, and a SQLite state store that remains under 80MB through a compaction pyramid, regardless of deployment duration. Production deployments require the SQLite state path to be placed on an encrypted filesystem. Typical use cases involve deploying Ori on edge devices to manage physical actuation tasks such as controlling motors, valves, or relays based on sensor inputs. The workflow begins with sensor events being processed through layers of reasoning, from deterministic rules to local SLMs, with each decision subject to the trust framework. Human correction of the machine's behavior permanently alters future actions, building trust over time. The runtime is currently in a stable 1.0.x channel, with related repositories in the organization covering additional components. Ori is designed for systems that must operate under constraint, making it suitable for industrial automation, remote monitoring, and other applications where reliable, autonomous physical action is required without constant cloud connectivity.