Approach · Layer 3.5

The layer between the model and the work.

Foundation models are Layer 3. A company’s real operations are the ground floor. The value nobody has captured is the seat in between — where a living model does the work.

What Layer 3.5 is

Layer 3.5 is not a replacement for your systems and not another chat box bolted onto them. It is an operating layer that runs on top of what you already have — a living model that plugs into a live process and takes it on end-to-end, with humans in the loop at the decisions that matter.

Because it runs on top rather than rips-and-replaces, a partner can measure the return before committing to anything larger.

Where we sit

The design-partner shape

The same simple deal across every proving ground.

01 You bring

  • Deep domain expertise
  • A real, live environment and its data
  • A scoped, paid pilot on one workflow

02 We bring

  • A living agent on your live process
  • Something static AI simply cannot do
  • Human approvals and full audit, built in

03 What stays whose

  • Your data stays yours — always
  • We keep the core model technology
  • The generalized capability carries to the next partner

Dual-track: open and proprietary

We open-source the generic framework for adoption and credibility, and gate the enterprise agent, the orchestration, and the proprietary, continuously-trained skills behind commercial licensing. Partners keep portability at the open layer and never get locked in where it counts.

Three proving grounds

Chosen by research yield — where a live, continuously-learning agent gets the richest feedback. We keep all three open and let the first well-fit partner set the lead.

01

Factory automation & robotics

Why it fits
Live sensor, vision and control streams; constant drift from wear and product-mix change; long-horizon line state; real-time control actions. The richest environment for the full stack.
First living-agent project
An always-on line agent that ingests the live stream, detects novel defects it was never pre-trained on, flags process drift, and predicts maintenance — escalating to a human.
The unlock
Catching new defect types in real time, with memory of the line’s entire history — exactly what a retrain-on-a-schedule model cannot do.
02

Finance & FinTech

Why it fits
Live transaction and market streams; fraud and risk patterns that shift by the hour; long-horizon per-entity state; consequential hold-flag-escalate actions. The founder is a fintech CTO — domain credibility is the wedge.
First living-agent project
A real-time risk agent on a live transaction stream that adapts to new patterns the moment they emerge, holds per-entity memory, and acts with human approval and full audit.
The unlock
Novel-pattern fraud caught in real time — without waiting for the next model retrain, the classic failure mode of static risk models.
03

Energy, battery & grid

Why it fits
Continuous telemetry; real-time optimization and dispatch; online degradation learning; long-horizon asset state — riding Japan’s green-tech and battery push.
First living-agent project
An always-on telemetry agent that learns online, per asset — battery-degradation prediction that adapts per cell, real-time dispatch under drift, or network anomaly detection.
The unlock
Online adaptation to drift and novel conditions in a safety-critical continuous system, with human oversight throughout.

From pilot to platform

A pilot proves one workflow and seeds the agent’s memory with your context. From there it converts to an annual engagement across the fleet, the lines, or the segment — and the capability we proved becomes a productized Layer 3.5 offering for the next partner.

Start a conversation