bluefabric
Sign in Book a Demo →
bluefabric/Architecture
system architecture

A four-layer fabric for supply chain AI.

bluefabric sits between your fragmented operational systems and the AI agents that need to read, reason, and act on top of them. Four layers, one job: turn messy, disconnected supply chain data into a clean, secure foundation that any AI can use safely.

// system architecture
L4 · CONSUMERS

AI Agents & Copilots

ClaudeChatGPTMicrosoft Copilot PerplexityGeminiCustom enterprise agents Open-source modelsWorkflow automation
L3 · INTERFACE

MCP layer + governed write-back

300+ read methodsanalyticscalculations recommendationssafe write-backaction routing approvalsaudit logs
L2 · FABRIC

The bluefabric intelligence fabric

connectorscleaningentity matching enrichmentcommon data modeloperational context supply chain knowledgegraph relationships
L1 · SOURCES

Existing operational systems

WMSTMSERPOMS EDIExcel / CSVAPIs databasesdata lakeslakehouses legacy systemssupplier & customer portals
Layer 1 · sources & connectors

Connect to the systems your operation already runs.

bluefabric is built for the reality of enterprise supply chains: data scattered across operational, analytical, and legacy environments, in formats nobody chose. Keep your systems. Add the intelligence layer above what already exists.

ERP

SAP S/4HANA, Oracle, Microsoft D365, NetSuite, Infor, and mid-market systems.

WMS

Manhattan, Blue Yonder, Körber, Softeon, Deposco, and custom WMS platforms.

TMS & Visibility

TMS platforms, carrier systems, project44, FourKites, and transportation feeds.

Files & Data Platforms

Excel, CSV, SFTP, Snowflake, Databricks, BigQuery, Redshift, APIs, databases, data lakes, and lakehouses.

Ingestion modes

  • Real-time event streaming — for inventory, shipment, and exception updates
  • Scheduled pulls — for ERP master data and finance close cycles
  • Change-data-capture — for high-volume transactional tables
  • Webhook & trigger-based — for partner and customer portal events
  • File-based ingest with schema inference — for the spreadsheet long tail
  • Backfill & replay — for historical context and model warm-up
Layer 2 · clean & model

Clean, match, enrich, and model your supply chain data.

Raw data from WMS, TMS, ERP, and trading partners arrives fragmented, inconsistent, and incomplete. L2 resolves it into a connected supply chain graph — cleaning duplicates, matching entities across systems, enriching missing attributes, and inferring relationships no single system explicitly records.

The graph is built on 16 supply chain entity and signal types, defined from 1,000+ consulting engagements. Explore the Data Model →

Cleaning & matching

  • Duplicate record resolution across systems with conflicting keys
  • Field-name reconciliation across WMS, ERP, and trading partners
  • Format harmonization — UoMs, dates, currencies, SKU codes
  • Entity matching — same SKU, supplier, warehouse, or customer across sources
  • Missing relationship inference — orders↔shipments, POs↔ASNs, lots↔batches
  • Deterministic + probabilistic matching with audit trail

Enrichment & context

  • Derived metrics — service-level risk, on-time-in-full, fill rate, dwell time
  • Operational state — what is late, stuck, at risk, or trending out of bounds
  • Cross-system flow — inbound delays mapped to outbound exposure
  • Historical context — exceptions, root causes, prior interventions
  • Supply chain knowledge graph — best practices, codified domain expertise
  • Source-of-truth resolution — which system owns which fact
Layer 3 · MCP intelligence layer

Trusted methods, exposed as MCP.

Once data is connected, cleaned, and modeled, bluefabric exposes hundreds of structured supply chain methods through the Model Context Protocol. AI agents call methods they can trust instead of guessing from raw exports — making AI faster, cheaper, more accurate, and easier to govern.

Method categories

  • Inventory & availability — by SKU, location, lot, status
  • Order status & fulfillment — open, in-flight, exception, completed
  • Shipment & transportation — risk, ETA, dwell, lane performance
  • Supplier performance — OTIF, lead time, quality, score
  • Warehouse capacity — labor, equipment, slotting, dock
  • Exceptions — detection, classification, root cause, suggested action
  • Cost & service — drift detection, anomaly explanation
  • Demand & commitments — customer commitment checks, what-if
mcp · bluefabric.supply_chain
// Agent calls a structured method instead of guessing
await mcp.call("bluefabric.getShipmentRisk", {
  shipment_id: "SHP-08821-A",
  include: ["weather", "carrier_health", "customer_sla"]
})

// → returns structured, governed response
{
  risk_score: 0.74,
  primary_driver: "carrier_capacity",
  eta_window: ["2026-05-09T14:00Z", "2026-05-10T02:00Z"],
  customer_commitment: "2026-05-09T17:00Z",
  suggested_actions: [
    "reroute_via_alt_carrier",
    "notify_account_manager",
    "escalate_to_priority_dispatch"
  ],
  data_freshness: "2m ago",
  source_of_truth: ["TMS", "WMS", "weather_signal"]
}
Layer 3.5 · governed write-back

Governed write-back for AI agents.

Most AI systems stop at read-only access. bluefabric lets agents safely act. Every write-back request is validated, permissioned, routed, and logged before it reaches a source system. Your WMS, TMS, and ERP never see raw AI output.

01
Intent

Agent submits a structured action request

02
Validate

Schema check, business rule + permission validation

03
Approve

Auto-approve or route to a human depending on policy

04
Route

Format and dispatch to correct WMS / TMS / ERP / OMS

05
Audit

Log full trail: who, what, why, where, when

Deployment & security

Secure deployment for critical supply chain data.

bluefabric is SOC 2 certified and designed for secure deployment in environments where operational data is sensitive and business-critical. Built for the way supply chain teams actually deploy AI infrastructure.

// option 01

bluefabric Dedicated Cloud

A managed, isolated bluefabric environment dedicated to your tenant. Single-tenant compute and storage, encryption at rest and in transit, controlled access, and SOC 2 certified operations. The fastest path to production.

// option 02

Customer Cloud Deployment

Deploy bluefabric in your own AWS, Azure, or GCP environment so sensitive operational data never leaves your perimeter. Same product, same MCP layer, same governed write-back — running on infrastructure your security team already owns.

Security & governance

  • SOC 2 certified operations
  • Controlled access with role and attribute-based policy
  • Isolated customer environments — no shared data planes
  • Auditable activity for every read, write, and agent action
  • Governed write-back — no direct AI access to source systems
  • Encryption at rest and in transit, key management options

Why this matters for AI

  • Your sensitive operational data is never used to train external models
  • Agent actions cannot bypass your existing approval workflows
  • Every AI-driven write is reversible and explainable
  • Source-of-truth boundaries are enforced — agents cannot accidentally corrupt master data
  • Procurement, security, and IT can review one product, not 12 integrations
  • Built so your CIO's first review is the last review

Book a technical demo.

30-minute walkthrough of the architecture, the MCP layer, and a live governed write-back action against a sandbox WMS.

Book a Technical Demo → Back to overview