Shipsy Logistics AI · Technical overview
← Back to event
Shipsy Logistics AI · technical overview

System of Action
over System of Record.

Your TMS records. Your WMS records. Your ERP records. Shipsy acts.

Vertical AI On any SOR Hot start · 4–8 wk pilots SAP-native
↓60%Manual ops firefighting
↓40%Customer WISMO queries
120h → 2hDispute resolution
4–8wPilot to live
20+SOR connectors
The category

Logistics has been recording for forty years.
It's time it started acting.

System of Record

Watches.
Reports. Waits.

  • × Logs shipments after they happen
  • × Surfaces exceptions on a dashboard
  • × Generates reports nobody reads on Monday
  • × Waits for a human to do something
System of Action

Detects.
Decides. Acts.

  • → Reads context across every system you run
  • → Detects the exception in <90 seconds
  • → Reroutes, reconciles, replies, replans
  • → Only escalates the edges that need judgement

Shipsy does not replace your TMS, WMS or ERP. It reads, reasons, and acts on top of any existing SOR — then writes the decision back. No migration. No rip-and-replace.

Integration · the picture

Your stack, untouched. One Shipsy layer on top.

All your existing systems become inputs to a single execution layer.

What you get
Autonomous routing Auto-reconciled invoices Proactive customer comms Resolved disputes
Shipsy · System of Action
Atlas Clara Nexa Vera Astra
Ontology · Reasoning · Connectors
Your systems of record (each customer brings their own)
SAP S/4HANA Oracle TMS Manhattan WMS Microsoft D365 Roadnet / iTrack Custom · in-house

Every customer arrives with a different stack. The Shipsy layer is the same; the connector layer translates. No migration, no rip-and-replace.

Architecture · the layers

Four engineered layers.

What sits inside the Shipsy layer.

01 Agent layer

AgentFleet — autonomous logistics agents

Each agent is a first-class microservice with its own identity, RBAC, audit trail, and bounded autonomy. Identities map 1:1 to a human operator in your SOR — same permissions, lower blast radius.

Atlas · Control TowerClara · SettlementsNexa · CX Vera · InboundAstra · DriverOpsAgent Builder · custom
02 Reasoning layer

Memory engine + multi-model routing

Session memory, long-term pattern memory, tribal-knowledge SOPs, and a multi-LLM router that picks the right model for each task — long-context planning vs. fast classification vs. structured tool-call.

Session memoryPattern memorySOP knowledge base Tribal-knowledge extractorMulti-model routerEval + drift detection
03 Ontology layer

Logistics knowledge graph + data lake

Transforms any SOR data model into a unified logistics schema. Auto-resolves master-data conflicts before agents run — so a shipment that has three IDs across TMS, WMS, and the carrier portal is correctly recognised as one.

Shipment schemaDriver entityTariff structure SLA contractsException taxonomyMaster-data normaliser
04 Connector layer

20+ pre-built integrations · supported protocols only

A multi-protocol ingest fabric. Each agent runs with scoped credentials; RBAC mirrors the human operator it replaces. Idempotent, replayable, lineage-tracked on every event.

SAP-native connectorOracle TMSManhattan WMS Roadnet · iTrackCarrier APIsREST · SOAP · EDI · EDIFACT · SFTP
SOR Your system of record

Untouched. Until you choose otherwise.

Shipsy sits on top — no migration, no rip-and-replace. Path A integrates with Shipsy TMS as the SOR; Path B with any third-party TMS, WMS, or ERP; Path C is standalone via the Agent Builder.

Three deployment paths · same agent layer

Wherever your data lives, we can act on it.

Path A

Shipsy TMS as SOR

Deepest integration, fastest setup. Real-time bidirectional data sync. All five AgentFleet modules available out of the box.

  • Real-time data sync
  • All AgentFleet modules available
  • Agents inherit RBAC natively
Path B

Third-party SOR
SAP · Oracle · Manhattan · custom

Twenty-plus pre-built connectors. SAP API restrictions bypassed via our native connector. Master data auto-resolved at onboarding.

  • 20+ pre-built connectors
  • SAP API restrictions bypassed
  • Master data auto-resolved
  • Same agent capabilities as Path A
Path C

Standalone Agent Builder

Low-code / no-code interface. Custom workflows beyond the five named agents. Bring your own data sources; full access to all layers above.

  • Low-code / no-code interface
  • Custom workflows beyond modules
  • Full access to all layers
Why Shipsy

Six moats horizontal AI cannot cross.

Structural advantages. Years to build inside a specific vertical.

01

Hot start, not cold start

Horizontal AI: 4–9 months before useful logistics output. Shipsy: pre-trained on logistics ontology, eval sets, and connectors. Agents arrive knowing consignments, SLA terms, exception codes.

Pilot results in 4–8 weeks vs 4–9 month cold-start ramp.
02

SAP-native bypass

SAP blocks horizontal AI vendors from core modules via API. Shipsy's logistics-native connector bypasses this entirely — years of integration work that is hard to replicate quickly.

Enterprise customers on SAP ERP with zero API restrictions.
03

Compounding intelligence

Every deployment enriches the shared logistics knowledge graph. Driver patterns from one customer inform another's anomaly detection. The n-th deployment is smarter than the first.

Compounding margin grows with each new customer.
04

Master-data standardisation

The same shipment with three different IDs across TMS, WMS, and carrier portal — resolved automatically at onboarding. Horizontal AI fails silently here; output looks slightly wrong, forever.

Real case: one part, 3 master codes across systems. Resolved automatically.
05

Mobile-first human-in-the-loop

Desktop HITL → hours of delay → agent blocks or takes an unsafe default. Push notification with rich context → one-tap approval. No response in 15 minutes? The agent calls the ops manager autonomously.

Zero workflow blocking due to HITL delays across all live deployments.
06

Three-tier safety architecture

Tier 1: pre-deploy eval sets, logistics + regulatory. Tier 2: real-time drift detection + circuit breaker. Tier 3: post-run pattern analysis + compliance audit trail.

Healthcare cold-chain: zero compliance incidents in 18 months of autonomous operation.
Safety & governance

Three-tier safety. Zero hallucinations at scale.

Calibrated for freight, healthcare, customs compliance — not content moderation.

Tier 1 · pre-deployment

Eval sets

  • Logistics-specific prompt eval suite
  • Regulatory scenarios — healthcare, hazmat, customs
  • Adversarial input testing
  • Bias detection across driver / region / carrier
  • No go-live without a passing score
Tier 2 · real-time

Drift detection

  • Continuous output distribution monitoring
  • Confidence-score floor enforcement
  • Anomalous-action rate thresholds
  • Cross-agent consistency checks
  • Auto circuit breaker → HITL on drift
Tier 3 · post-run

Pattern analysis

  • Decision outcome correlation analysis
  • Human override pattern mining
  • Regulatory audit trail generation
  • Agent-retraining trigger detection
  • Monthly compliance report
Models supported

Frontier providers, routed by task type + cost

No training on your data

Enterprise contracts with all providers prohibit training on customer inputs

PII minimisation

90% of logistics use cases need zero PII — stripped before the LLM boundary

Data residency

EU · APAC · US — data never leaves designated boundary

One workflow. Prove it.
Then expand everything.

Start wherever you have the most manual firefighting today — Control Tower, CX, or Settlements. We will tell you which one in the first session.

← Back to event Request a diagnostic →