Every door into the data bed goes through the same policy engine: REST for applications, SOAP with a WSDL generated from the model for older integrations, SQL through JDBC and ODBC for BI and analysts, events for consumers that sync, files for exchange, and an agent endpoint.
| Door | For | Read | Write |
|---|---|---|---|
| REST | applications, portals, SmartApps | GET with fields=, filter=, expand=, as_of=; pagination and limits published | PUT and PATCH on any combination; POST to create or propose |
| SOAP | older integrations and middleware | select operations generated from the model (WSDL) | insert and update operations, bulk or per record |
| SQL (JDBC · ODBC) | BI tools, analysts, spreadsheets | golden views and the bed, with AS OF | governed data entry through the workbench; bulk as a table |
| Events | consumers that sync | webhooks and Kafka topics per domain: created, changed, merged, retired | — |
| Files | exchange with partners and legacy systems | governed extracts with the security filters applied and logged | loads with the same rules as any other write |
| Agent endpoint (MCP) | copilots and agents | the same reads as REST, as the agent's identity | proposals, logged at the agent's level |
The fields parameter takes any list of attributes, nested groups and related objects: legal_name, vat_id, sites.address.postcode, open_pos.count. The same syntax works in REST, in the SOAP select operation and as a column list in SQL, so an integration written once reads the same shape from every door.
Bulk is a table: a bulk read is a SQL query or a paged GET; a bulk write is a load with the same rules as a single PUT. There is no separate job API with its own limits.
# three doors, one shape
GET /api/v1/golden/supplier?fields=legal_name,vat_id,sites.address.postcode,open_pos.count&filter=country:DE
SELECT legal_name, vat_id, sites.address.postcode, open_pos.count
FROM golden.supplier WHERE country = 'DE';
<sup:select><sup:fields>legal_name vat_id sites.address.postcode open_pos.count</sup:fields>
<sup:where>country=DE</sup:where></sup:select>Yes; the WSDL operations run through the same engine as REST and SQL.
Change events come from the ledger, so source-side changes read in place and hub-side edits both fire; a consumer can replay from any point.
Yes; an extract runs as the partner's identity, so masks and row filters apply, and the extract is logged.
Bring two systems that hold the same customers, suppliers or materials. In one working day we connect them, run the match on live rows, review the queue with your steward and publish the first golden view with its API.