Resemver: Using AI to rewrite semver based on actual breaking changes (fossa.com)

Resemver uses AI to automatically analyze code diffs and determine whether a change is a breaking API change that should bump a package’s major, minor, or patch version under semantic versioning (semver). Instead of relying on commit messages or human judgment, it combines program analysis (AST and signature diffing, call graph / type checks where available) with large language models trained on code to classify changes, generate suggested version bumps, and produce human-readable explanations and changelog entries. It can be hooked into CI/CD pipelines and pull-request workflows so version decisions are made as part of review and release automation. This matters because misapplied semver is a common source of dependency breakage and silent regressions across the open-source ecosystem. By surfacing concrete API-level differences (removed or renamed functions, changed parameter types/defaults, changed return shapes, altered thrown errors or side effects) and pairing them with LLM-driven rationale, Resemver promises faster, more consistent versioning and less manual overhead for maintainers. Key caveats: accuracy depends on test coverage, type information, and the quality of analysis models, so human review remains important; there are also provenance and privacy concerns when sending code to models. Overall, Resemver illustrates a practical use of AI+static analysis to reduce semantic drift in package ecosystems and improve release automation.