IonIONengine
Examples
GitHub
Examples

A distribution center with AI in the right places

Six deployments, six configurations, one binary

Receiving dock · inbound pipeline

The paperwork disagrees before the pallet moves

A trailer arrives with a partial pallet and a packing list that does not match the advance notice. The engine reads both, works down the open purchase order line by line, and posts what disagrees before the pallet is put away.
Which engine surface runs itA stage inside the inbound pipeline you already run.
Who sees the outputThe receiving lead, on the terminal already at the dock.
What it replacesComparing a printed packing list against a screen, then emailing purchasing and waiting.
This engine instanceRuns as · A container invoked as one pipeline stage, then it exitsModels · A local model for extraction; a hosted one only when lines disagreeExtensions · receiving.tsTriggered by · The pipeline, on check-in

The mechanics behind Receiving dock, in this example

One worked example, not a prescription. Receiving dock could use agents and schedules as well; this example did not need them.

A distribution center with AI in the right places

Six deployments, six configurations, one binary

The mechanics behind Receiving dock, in this example

One worked example, not a prescription. Receiving dock could use agents and schedules as well; this example did not need them.

Receiving dock · inbound pipeline

The paperwork disagrees before the pallet moves

A trailer arrives with a partial pallet and a packing list that does not match the advance notice. The engine reads both, works down the open purchase order line by line, and posts what disagrees before the pallet is put away.

Which engine surface runs it

A stage inside the inbound pipeline you already run.

Who sees the output

The receiving lead, on the terminal already at the dock.

What it replaces

Comparing a printed packing list against a screen, then emailing purchasing and waiting.

What the extension author wrote

  • +2 tools: readASN, openPOLines. Both read-only.
  • +No hooks. The ERP tools you handed it have no write path, so there is nothing to police.
  • +Trigger: one pipeline stage on trailer check-in

Everything else was already there: the model calls, the retries, the tool loop, the permission gate, the event stream.

This engine instance

Runs as
A container invoked as one pipeline stage, then it exits
Models
A local model for extraction; a hosted one only when lines disagree
Extensions
receiving.ts
Triggered by
The pipeline, on check-in
A person
No. It writes a note and moves on.

None of these six is the same deployment. Different containers, different model lists, different extensions, different limits. The same binary underneath each one. There is no central engine everything funnels through, because there does not need to be.