Skip to content
Art2link ESB HomeDocumentationBlogContact
Updated August 2, 2026
AI features/AI Accelerator

AI Accelerator

One free-form chat that builds three things: pipeline components, custom functions, and maps.

The Accelerator is a chat, nothing more structured than that. You describe what you need in your own words and it produces it. The prompt is the whole interface, which puts the weight on the prompt: what you state in it is what the Accelerator has to work from, and everything you leave out it has to guess.

The three artifact-specific articles below are about exactly that. None of them lists anything the platform demands of you; each one lists the facts that make the difference between a first draft you can use and one you have to rework, and shows a prompt that states them.

Industry standards, built in

The name is deliberate. The AI Accelerator is where domain knowledge lives, and that knowledge covers the message standards integration work actually runs on. It already understands the major EDI, healthcare, financial, supply-chain, and general-purpose formats, so you do not have to teach it a specification before you can put it to work. Ask it to disassemble an EDI 810, validate an EDIFACT ORDERS, parse an HL7 ADT feed, read a HIPAA 837 claim, or handle a SWIFT MT103, and it gets to work: no add-on, no package to license, no separate install.

Everything BizTalk shipped an accelerator for, and a great deal more, is already in the knowledge base:

StandardRepresentative transaction types
EDI (X12)850 purchase order, 855 PO acknowledgement, 860 PO change, 810 invoice, 856 advance ship notice, 846 inventory advice, 820 payment/remittance, 830 planning schedule, 940/945 warehouse shipping, 204/210/214 transportation, 997/999 acknowledgements, and the rest of the set.
EDIFACTORDERS, ORDRSP, DESADV, INVOIC, DELFOR, RECADV, IFTMIN, PRICAT, CONTRL, APERAK, and the wider UN/EDIFACT directory.
HL7Clinical and administrative messages such as ADT, ORM, ORU, SIU, MDM, DFT, and MFN, down to the segment and field level.
HIPAA (healthcare X12)837 claims, 835 remittance advice, 270/271 eligibility, 276/277 claim status, 278 authorization, 834 enrollment, 820 premium payment.
FHIRResources and bundles (Patient, Observation, Encounter, and the rest) for modern healthcare interoperability.
SWIFTMT messages such as MT103, MT202, MT940, and MT950, plus the ISO 20022 MX equivalents.
RosettaNetPIPs such as 3A4 purchase order, 3B2 advance shipment notice, and 4B2 shipment receipt.
ISO 20022Payment and cash-management messages, the pain, pacs, and camt families.
SAP IDocCommon IDoc types including ORDERS, INVOIC, DESADV, and DELVRY.
Flat files, CSV, JSON, XMLThe everyday formats too, fixed-width and delimited files, JSON, and XML.
The table is a starting point, not the boundary. The AI Accelerator’s model is aware of essentially every message standard in common use, well beyond the list above. If you work with something that is not shown here, do not assume it is unsupported, just ask. Prompt it directly, for example “What message standards and formats can you work with?” or “Can you disassemble an EDIFACT DESADV?”, and it will tell you what it can handle before you commit to a build.

If you are coming from BizTalk, this is where the accelerators went. On BizTalk you installed a separate accelerator for each domain, EDI, HL7, SWIFT, RosettaNet, and so on, before the engine knew how to handle those formats. In Art2link ESB that knowledge is embedded in the AI Accelerator’s knowledge base. The same request that used to require the right accelerator on the box is now just a prompt.

The three artifacts it builds
Ask it forWhat comes back
Pipeline componentA custom pipeline component, the C# class, its metadata, and its configurable properties.
Custom functionA custom function, the C# method, the [CustomFunction] attribute, and the call-site usage.
MapAn Art2link map, the XSLT, the source/destination shape alignment, and any helper-function calls.

Those are the three artifact kinds it generates. Everything else you type into the chat is a question it answers rather than an artifact it builds.

What makes the output better WORTH PUTTING IN THE PROMPTWhat to buildThe shapes involvedThe Message TypesThe edge casesAI Acceleratorfree-form chatPipeline componentCustom functionMapYou reviewand savethe same chat also answers questions and explains artifacts you already have
A prompt only works as well as what you put in it. The Accelerator knows the standards, the C# surface, and how Art2link ESB artifacts fit together; what it cannot know is the part that is specific to your integration, the field a value comes from, the boundary a batch splits on, the Message Type to stamp on what it emits. State those and the first draft usually lands close. Leave them out and it fills the gap with a reasonable guess, which compiles and then does the wrong thing at runtime.
Beyond the three artifacts

Because it is one open chat, generating an artifact is only part of what you can do with it. It is also the fastest way to:

  • Ask what a standard contains before you build against it, which segment carries a value, what a transaction set looks like.
  • Work out why an artifact you already have behaves the way it does, for example why a map is producing an empty BuyerName.
  • Ask for several artifacts in one request, a custom function and the map that calls it, and have them line up with each other.
  • Talk through naming or design before any code gets generated.
What the Accelerator can’t do yet
⚠️
The Accelerator cannot build an integration end-to-end. It can generate the individual artifacts described above, pipeline components, custom functions, maps, but it cannot today wire up a receive port, a routing rule, a map, and a send port into a working flow on your behalf. Whole-integration generation is in development, with no release date.
Availability
Included in all editions.