Claryo

Claryo develops a "warehouse world model" and a set of AI agents intended to operate on top of it. The model is described as pre-trained across real warehouse operations and then fine-tuned to an individual facility, using footage from cameras already installed on site fused with warehouse management system (WMS) transactions. A short scan grounds the model in the building's geometry and mapped WMS locations, after which live data keeps it aligned with current conditions. The company frames the model as holding a facility in three tenses: replaying past operations for query and incident audit, mirroring the present for real-time prediction of backlogs, bottlenecks and work stoppages, and simulating future scenarios such as volume spikes and layout changes in a spatially accurate 3D environment.

On top of the model, Claryo offers six AI agents covering distinct warehouse workflows: Production Management, Shift Planning, Continuous Improvement, Inventory Audit, Safety & Compliance, and Reporting. The agents are positioned along a five-level autonomy scale modeled on autonomous-vehicle terminology — inform, predict, recommend, act with notification, and autonomous resolution of a defined workflow within guardrails — with the customer setting the level per workflow and retaining override, governance and escalation. Functionally the product spans three application areas: monitoring (real-time visibility across zones, processes, labor and inventory), planning (shift-by-shift and seasonal what-if analysis of layouts, labor, robot fleets and peak volumes), and orchestration (dispatching tasks and allocating labor, equipment and robots, and re-planning around exceptions).

A second product line targets robotics vendors, using a simulation of the deployment site instantiated from that facility's world model. Vendors can validate ROI before installation, commission robots virtually before going live, and evaluate navigation, perception and fleet behavior under varying conditions. In production, Claryo's agents localize robots against the building using ceiling cameras, cross-checked against floor geometry, and monitor fleet performance and SLA compliance. The company states the simulation works with any robot stack and generates the sensor streams robots consume. A third audience referenced by the company is robotics frontier labs, to which it offers training data on the facility's terms.