Boomi Details How Scribe Turns Integration XML Into AI Documentation

Boomi has detailed the architecture behind Scribe, an agent that generates documentation for integration processes as developers work. It converts integration XML into a workflow graph for Claude Haiku 4.5 to describe, then compares current and previous graphs to flag changes for developers.

One process map, two outputs

Boomi integration processes begin as XML files. Scribe extracts their features and converts them into Directed Acyclic Graph, or DAG, dot notation, mapping connected workflow steps. Boomi sends that structured input through Amazon Bedrock to Anthropic’s Claude Haiku 4.5 to generate documentation.

Those workflows govern how data is retrieved, manipulated, routed, and delivered between source and destination systems. Boomi specifies documentation that begins with an objective and process representation, then covers metadata, business context, and step-by-step functions. Metadata can include the process name, version, dates, path, and number of components.

Version review takes a separate path: Scribe compares the current DAG with a previous version to identify additions, modifications, deletions, and observations developers can act on. Boomi characterizes the earlier manual documentation and comparison work as time-consuming and error-prone, with incomplete or inaccurate descriptions contributing to misunderstandings and rework.

Boomi also says complete documentation can help prevent gaps in auditing for business and compliance needs. The aim is a common record of workflows that may otherwise be difficult for anyone other than the builder to understand.

A pipeline around the model

The language model is one component of the stack. Boomi uses Amazon SageMaker AI for user-intent classification models; AWS Lambda orchestrates parsing, documentation generation, and comparison. Amazon S3 stores DAG files, generated documentation, and metadata, while DynamoDB serves as the internal backend datastore.

A DAG uploaded to S3 triggers Lambda parsing of its nodes and edges. The parsed graph is passed to the model for documentation, and the results are stored in S3; the Lambda layer also handles comparison of current and earlier graphs. That separation assigns generation, intent classification, storage, and orchestration to distinct services rather than to the documentation model alone.

Documentation in the product interface

Generated documentation is available in the Boomi Integration Canvas and Boomi GPT. The architecture is intended to support developers across more than 33,000 Boomi customers, which describes its design scale rather than confirming adoption by every customer.