Skip to content

Platform / ERP Context Layer

Make your ERP agent-ready.

SAP S/4HANA, Oracle NetSuite, Microsoft Dynamics, Workday, IFS. Whatever you run, your custom fields and your approval chains stop being a black box. They become context an AI agent can act on without touching production.

Sandbox connection provisioned by your IT team. Scoped to the authorizations you approve. Audit log on every call.

before · SAP GUI

ME23N

EKKO-EBELN    4500002847
EKKO-LIFNR    0000100482
EKKO-WAERS    EUR
EKKO-BUKRS    1000
EKKO-ZTERM    NT45
EKPO-MATNR    000000000087413
EKPO-MENGE    12 000,000
EKPO-NETPR    3,42
EKPO-EINDT    20.05.2026
EKPO-WERKS    1000
EKPO-LGORT    0001
EKPO-ZZRISK   B2
EKPO-ZZAUDIT  Y
*ZAPPROVAL_CHAIN: U-EISIK01;
                   U-HAKAN;...

after · MCP tool

agent-ready
{
  "tool": "procurement.po.read",
  "input": { "po_id": "PO-2847" },
  "schema": {
    "supplier":     "Steelcore Ltd",
    "currency":     "EUR",
    "payment_term": "Net 45",
    "items": [{
       "sku":          "STL-8741",
       "quantity":     12000,
       "unit_price":   3.42,
       "delivery":     "2026-05-20",
       "plant":        "Bursa 1",
       "risk_flag":    "watch",
       "audit_locked": true
    }],
    "approval_chain":
       ["emre.ik", "hakan"]
  },
  "principal": "emre.ik@spaceflow.tech"
}

one ERP table · one MCP tool · zero custom integration code

4,182 fields mapped

What this replaces

Your ERP is the source of truth. It just doesn't speak agent.

Three realities of putting AI on top of an enterprise system. None of them are about the model.

Custom fields and Z-tables nobody documents

~$2M/yr

Every ERP install is a snowflake. Custom tables hold the business logic, the workarounds, the institutional knowledge. Off-the-shelf AI sees none of it and answers with the standard schema, which is the wrong schema. The mapping work is what we do first.

Approval chains live in the role layer

Compliance risk

Who can approve what, up to what value, under which delegation, on which company code. None of it is in the data layer. Agents that don't read this end up either gated by humans for everything, or worse, escalating things they shouldn't.

Every project starts integration from zero

6-month rollouts

RFC connections, BAPI wrappers, IDoc handlers, OData services, NetSuite SuiteScript, Dynamics OData. Every new use case rebuilds the same plumbing. By the time you've shipped one workflow, the model you picked is two generations old.

How a connection actually proceeds.

  1. Phase 01

    Sandbox access provisioned by your team

    Your BASIS or IT team provisions a scoped service user in your sandbox tenant with the BAPI and OData authorizations you approve. Spaceflow operates only inside those authorizations. Every call appears in your standard ERP audit. Production is untouched at this stage.

  2. Phase 02

    Mapping session against the sandbox

    Working session with your team. We walk through your data dictionary, your custom append structures, your Z-tables, and the relationships between them. You correct anything that does not match how your business actually uses the system. Output is a reviewed map your team signs off on.

  3. Phase 03

    Expose mapped objects as MCP tools

    Each mapped object becomes a typed endpoint that any AI agent (Claude, Copilot, your own) can call. Authorization stays enforced by your existing PFCG roles. Tool definitions are versioned in your workspace and reviewed alongside any other production change.

  4. Phase 04

    Production cutover on your schedule

    Your team provisions the equivalent service user in production under the same authorization scope. The reviewed map, the typed tools, and the audit chain carry across. Go-live timing is yours to set.

Works with

Works with what you're already running.

SAP S/4HANASAP ECC 6.0Oracle NetSuiteOracle E-BusinessOracle FusionMicrosoft Dynamics 365WorkdayIFS CloudInfor M3JD EdwardsSage X3Epicor