Inferrex developer guide
This guide shows a developer how to use the Inferrex platform end to end: get access, connect a system, review its objects, map it, build a flow, review Golden records, and watch what runs. It then covers the same work from InferrexAPI, from InferrexMCP, and from InferrexCLI, the cloud shell.
The model in one screen
Everything in Inferrex follows one sequence:
System -> Object -> Mapping -> Flow -> Destination -> Watch
| Step | What it is |
|---|---|
| System | Something you connect by the address of its API. One connection per system, keyed by its own id. A provider with several organisations gets one connection per organisation. |
| Object | A kind of record the system holds, such as a customer or an order, with its fields. |
| Mapping | Which objects and fields in one system line up with another. Versioned and approved. |
| Flow | The thing that moves or changes data: a source, a mapping per destination, a trigger and a direction. |
| Destination | Where a flow sends records: another system, the Golden record, or a program. |
| Watch | Runs, failed writes, repairs, health and alerts for a system or a flow. |
Two rules shape everything else:
- Functional is a Flow, informational is a Report. If running it can change a record anywhere, it is a flow. If it only reads and tells someone, it is a report.
- Nothing is saved or shown as working until a live signed-in read passes. A system is not approved until Inferrex has read real data from it with your sign-in.
Words in the app and names in the API
The app says System and Flow. The API paths and MCP tool names still say connection and pipeline. They are the same things.
| In the app | InferrexAPI path | InferrexMCP tool names start with |
|---|---|---|
| System | /v1/connections |
connection_, connections_ |
| Sign-in and secrets | /v1/connections/:id/credentials, /v1/secrets |
credential_, credentials_ |
| Object and field | /v1/connections/:id/manifest |
connection_manifest |
| Mapping | /v1/mappings |
mappings_ |
| Flow | /v1/pipelines |
pipeline_, pipelines_ |
| Further destination | /v1/pipeline-targets |
pipelines_pipeline_target |
| Golden record | /v1/unified/records, /v1/golden-records |
unified_api_, golden_record_ |
| Golden-record change | /v1/canonical-proposals |
golden_record_ |
| Failed write | /v1/quarantine |
pipelines_list_quarantine |
| Safety stop | /v1/circuit-breakers |
monitoring_ |
| Repair | /v1/heals |
heals_ |
| Automation | /v1/workflows |
workflows_ |
| Report | /v1/reports/query |
reports_query_report |
| Environment | /v1/workspace/environments |
workspace_environments_ |
| Background job | /v1/jobs |
jobs_ |
Chapters
- Get started: a workspace, an API key, a first call.
- Connect a system: address, sign-in, live test, first read, approve.
- Objects and mappings: review what was found, match two systems, approve field links.
- Flows: create, dry run, turn on, sync, schedule, destinations.
- Golden records and review: read Golden records, decide proposed changes, work through Review.
- Watch and health: runs, failed writes, repairs, safety stops, alerts, reports.
- Environments and promotion: work in one environment, promote a flow to the next.
- InferrexAPI: authentication, scopes, idempotency, jobs, errors, outbound webhooks.
- InferrexMCP: the same operations as tools for AI coding tools.
- InferrexCLI, the cloud shell: a terminal for sign-ins that need a command-line tool.
Skills for AI coding tools are on the skills page. Each is one SKILL.md file.
Conventions in this guide
https://api.example.com/v2is a stand-in for the address of the system you are connecting.$INFERREX_API_KEYis your Inferrex API key, held in an environment variable. Never paste a key into source code or a chat.CONNECTION_ID,PIPELINE_ID,MAPPING_IDand similar are placeholders for ids returned by earlier calls.- Response shapes list the fields the API returns. They are shapes, not sample data.