AI assistant — Pipeline component
A guided prompt that collects the inputs a custom pipeline component needs, then generates the component code — so nothing about the component’s behaviour is left for the AI to guess.
Pipeline components are the most fact-dependent thing the assistant builds: the inbound message shape, the boundary that splits one message from the next, the promoted properties the downstream pipeline expects — each of these has to be exact, and getting any of them wrong produces a component that compiles but does the wrong thing at runtime. The guided mode exists to make sure the assistant has all of it before it writes a line of code.
Today the guided flow is centred on the disassembler workflow, since that is the component type users most often need help building from scratch. Additional component types will be added to the guided list over time; for component types not yet covered, use the free-form mode.
The form asks for the facts the assistant cannot infer:
- The inbound message shape (sample, schema, or description).
- Where one message ends and the next begins — the boundary the disassembler should split on.
- Any promoted properties the downstream pipeline, message-type routing, or send port expects, and where on the message they come from.
- Configurable component properties — values you want exposed in the Art2link ESB UI rather than hardcoded.
- Any edge-case behaviour the component must handle (e.g., empty batches, malformed records, encoding quirks).
Anything the form doesn’t collect, the assistant has no opinion on — that’s the point.