Amazon-like connector onboarding for an operator stack

Why integrations usually stall

An operator runs a stack of systems: dispatch, fleet, brokers, rewards, directory. Connecting them is where most platform value is promised and where most of it is lost. Each new integration becomes a bespoke project, with its own onboarding flow, its own monitoring, and its own way of failing quietly.

The Platform integration substrate is designed to answer that pattern. It is boring infrastructure by design: a single catalog to browse and enable connectors, every change auditable, every conflict visible. It is not a dispatch engine. It is the plumbing meant to make the rest of the suite cohere.

Onboarding as a side effect of the schema

Every connector is designed around the same shape: a manifest, an adapter, and a health probe, so the catalog page renders from the manifest registry instead of being hand-built per connector. Today that gets an operator a real catalog experience and a read-only view of what a connector would bring in. Two-way sync, live data pulled on a schedule and changes written back out, is the next build phase, not yet turned on.

That is what makes the intended experience feel like adding an item to a cart: an operator browses connector cards, enables a few, and reviews what a connector would surface, no developer flow required. The Amazon-like onboarding people ask for is a property of the schema we are building toward, not a bespoke feature written again for every system.

Reliable by construction

The plan underneath the catalog is a sync spine built in the right order: a transactional outbox, a durable queue, and a pool of idempotent workers, so that once live, no event is lost in a crash and no retry double-applies. Health is designed as a first-class metric from the start: a gateway probe plus an external monitor that second-guesses it, so degradation would surface on the same card the operator enabled, with a specific reason, and reconciliation diffs would catch drift before it goes unnoticed.

None of this is a promise to route around the hard part of integration work. It is a description of what we are building so that turning on a connector stays boring, instead of becoming a project of its own.