OpexAI Platform v0.4 · Public beta

An AI platform built to expand operating margin in mid-market business.

OpexAI is a managed platform and an embedded build team — three layers (ontology, agents, workflows) deployed into the systems you already run, observable end-to-end, owned by you.

03 Platform layers · ontology · agents · workflows
1× Embedded build team per engagement
0 Systems of record replaced
100% Decisions logged & auditable
In production at
GraniteTop + engagements under NDA
§ 01Architecture

A three-layer platform. Each layer depends on the one beneath it. Every customer engagement uses all three.

Three layers. One platform.

Decisions surface to your team
Layer 03 / 03 Workflows Surface HUMAN-IN-THE-LOOP

Where the right person sees the right decision at the right time.

Human-in-the-loop interfaces and decision moments. Designed around how work actually happens — owners, project managers, fabricators, schedulers, installers, finance. The right context surfaced from the ontology, the right recommendation generated by the relevant agent.

Designed around the operator, not the dashboard.
▣ Live decision queue · sample
Scheduler · Maria K. → approve job re-sequence 4m
Fabricator · Bay 02 → confirm slab swap 12m
PM · J. Patel → accept revised quote 28m
Finance · Controller → release vendor batch 1h
Layer 02 / 03 Agents Reasoning SCOPED · OBSERVABLE

Software entities with a specific operational mandate.

An agent has access to relevant slices of the ontology, the right model behind it, and guardrails appropriate to its scope. Agents are not chatbots — they observe the business, propose actions, and where authorized, take action. Always within scope. Always observable.

Catalog defined during Discovery, per customer.
▣ Active agents · GraniteTop catalog
pricing-margin Re-prices jobs · margin floor LIVE
slab-yield Optimizes cuts · waste < 8% LIVE
scheduler Sequences bays & installs LIVE
vendor-recon Reconciles invoices to POs LIVE
Layer 01 / 03 Ontology Foundation SINGLE COHERENT MODEL

A structured, living digital model of your business.

A single coherent representation of the things your business cares about — for an operating company, that universe includes inventory, jobs, machines, people, customers, schedules, costs, and outcomes — and the relationships among them. Sourced from your existing systems of record. We don't replace them; we ingest, harmonize, and present a single connected view.

Without it, agents make decisions on fragments. With it, they make decisions in context.
▣ Entity model · sample
JOBS CUSTOMERS SCHEDULES OUTCOMES SLABS MACHINES PEOPLE COSTS VENDORS INVOICES MARGIN
Sourced from your systems of record · ERP · CRM · MES · finance
§ 02The product

A console where the platform meets the operator. Margin metrics, agent activity, decisions in queue.

What it looks like in operation.

opex.ai/console — granitetop / production Live · synced 14s ago
Production / Console / Margin overview

Margin overview · trailing 90 days

90D Q2 Live
+412bps +1.8 pts vs. prior 90d Annualized · $2.1M operating-margin recovered to date
+5% +2.5% 0

Active agents 11 / 11 healthy

01 pricing-margin · re-priced 24 jobs RUN 412 ms
02 slab-yield · 7.6% waste · floor 8% RUN 8.2 s
03 scheduler · 18 bays · 4 installs RUN 221 ms
04 vendor-recon · 142 invoices RUN 1.4 s
05 install-route · 11 trucks IDLE

Decisions in queue 14 awaiting

Re-sequence jobs 8814 · 8819 · 8821 PEND M. K.
Approve slab swap · Bay 02 · 32sqft PEND Bay 02
Accept revised quote · #Q-22918 PEND J. P.
Release vendor batch · 04-2026-Q2 PEND Ctrl.
Approved · install-route · 04 · 7am DONE 2h ago
§ 03Delivery model

Software alone doesn't ship operating margin. People do — paired with the right platform.

A managed platform. Plus an embedded build team.

▣ Component A · The platform

Three layers, shipped as a managed service.

OpexAI runs as a managed platform. We host it, we operate it, we monitor it. Your team consumes it through the console and through the workflows we wire into your operations.

Hosted runtimeregional residency available
End-to-end observabilityEvery decision logged · full audit trail
Guardrails & scope controlPer-agent permission boundaries
Versioned ontologySchema migrations without downtime
▣ Component B · The embedded team

Senior engineers inside your operation.

Every engagement comes with an embedded build team — the people who design your ontology, scope your agents, and wire the workflows into how your business actually runs. They sit with your team during Discovery and stay through launch.

Discovery sprintTwo weeks · ontology & agent catalog defined
Senior on the keyboardThe people who design also build
Launch in weeksFirst agents live within 30 days · standard
Knowledge handed backYour team owns the operating runbook
§ 04Customer

A single coherent model of a fabrication business — and the agents that run on top of it.

GraniteTop. A fabrication operation, instrumented end to end.

GraniteTop runs a multi-bay fabrication operation with vendor-managed inventory, scheduled installs, and the kind of margin pressure familiar to every mid-market fabricator: everything matters, nothing is centralized.

OpexAI ingested from their existing ERP, MES, and finance stack — leaving every system of record exactly where it was. The ontology became the single coherent view of inventory, jobs, machines, people, schedules, costs, and outcomes. Agents were scoped during Discovery: pricing-margin, slab-yield, scheduler, vendor-recon, and seven more. Workflows were wired into the actual humans who run the business.

★ One workflow, end-to-end

  1. Ontology observes a slab is rejected mid-cut. Inventory, job, customer, schedule are all linked.
  2. slab-yield agent proposes a substitute slab from current stock that meets the customer's veining spec.
  3. scheduler agent re-sequences downstream installs so the swap doesn't push delivery dates.
  4. Workflow surfaces a single approval to the bay fabricator with one click — or one call.
  5. Audit log records the full chain: observation → proposal → approval → outcome. Reviewed weekly.
§ 05Brand identity

The system: a stacked mark, a mono wordmark, two accents, and three faces of type.

The system behind the surface.

The mark · The Stack

Primary
Mono
Reversed

The wordmark · opex.ai

opex.ai
opex.ai Horizontal
opex.ai Stacked
Surface · Ink
Ink
#0A0E0C 10 · 14 · 12
Surface · Lifted
Surface
#161E1A 22 · 30 · 26
Accent · Brand
Ember
#D6592B 214 · 89 · 43
Accent · System
Phosphor
#5BE3A8 91 · 227 · 168
Type · Primary
Bone
#EAE3D6 234 · 227 · 214
Type · Muted
Steel
#4D5A60 77 · 90 · 96
Display + UIGeist
Aa Gg Mm 0123 Headlines · navigation · body · buttons. Tracking −0.04em at scale.
Chrome + SpecJetBrains Mono
opex.ai Wordmark · labels · code · technical chrome. Uppercase 0.18em for labels.
Editorial AccentFraunces
expand Reserved for emphasis & editorial accent only.
§ 06 — Get on the platform

Ready to put the three layers to work? Tell us about your operation.

★ Beta access · 48h response · NDA on request