Six layers, end to end. From a worker's first-person capture in the field, through structured task graphs and outcome links, to deployed applications across robotics, insurance, copilots, and autonomous work systems.
Every signal flows up. Every prediction flows down. Layer 5 — the Eru Physical World Model — is the gravitational center.
Raw signal from the physical world. First-person POV, third-person context, site sensors, partner streams.
Signal becomes substrate. Synchronized, filtered, permissioned, ready for structure.
Pixels become a graph. Tasks, tools, objects, zones, sequences — all linked into an event graph.
The structured graph is bound to business truth. Inspections, rework, claims — Eru's source of causal signal.
The center of gravity. A 2.4B-parameter foundation model with perception, workflow, outcome, action, memory, and prediction heads.
Surfaces partners and customers actually use. PIN OS, insurance, robotics APIs, copilots, simulation, autonomous work.
The stack carries three distinct kinds of traffic, on the same six layers. Each flow has its own latency, retention, and access policy.
Worker captures → events structured → outcomes linked → model updates. Latency: minutes to days. Permanence: indefinite, permissioned, revocable.
Model heads → APIs → PIN OS dispatch, copilots, robotics partners, insurer dashboards. Latency: 12–80ms. Permanence: query-time only.
Outcome layer → PIN Certified → portable records that follow workers, machines, companies, and assets. Verifiable, signed, audited.
Layer 6 is where the model meets the world. Some are built by PIN. Some are built by partners. All draw from the same model.
Operational system for physical work. Scheduling, dispatch, supervision, inspection.
Portable proof-of-work records for people, machines, and assets.
Verified worksite signal feeding underwriting, prevention, and claims defense.
Task priors, action knowledge, evaluation, and handoff for partner platforms.
Human upskilling driven by expert demonstration and supervisor correction.
In-line guidance for trades, foremen, and shop-floor operators.
Long-tail scenario generation mined from real work for training and stress testing.
Mixed human-robot workflows orchestrated against verified outcomes.