gixsy95/abap_wiki

abap_wiki is a local, agent driven engine that turns a company's SAP S/4HANA system into a navigable, versioned knowledge base written in Markdown and viewable in Obsidian. It is aimed at the problem that in most real SAP systems, documentation is missing or scattered, and the knowledge of why something was built a certain way often only lived in the head of a developer who has since left. The tool starts at a company's custom SAP objects and follows their connections outward into the standard objects they touch, documenting each one in the context of how it is actually used. It builds this up in layers: first a basic inventory, then code analysis, then a functional and process level explanation. For the deeper layer, a human who owns the business process fills in gaps the AI agent could not figure out on its own, and the system only raises the documentation level after checking every answer against the actual code. A key feature is that every claim in a generated page is tied to specific lines of source code that prove it, and a separate agent independently rechecks the analysis and rejects anything it cannot confirm. If something cannot be proven, it is labeled as such rather than hidden or guessed at. The whole system runs entirely on the user's own machine, with no data sent to outside services, and it can be driven by Claude, Codex, or a local open source AI agent. The project frames itself as different from typical retrieval based AI search, which re-reads raw code every time a question is asked. Instead, abap_wiki builds the understanding once, verifies it, and lets future questions navigate an already checked set of pages instead of searching raw code again each time. It is built in Python 3.11 or newer, includes a command line tool for measuring token savings compared to raw code, and is released under the MIT license.