sem is a semantic version control tool that works on top of Git.
sem solves the problem that traditional line-based diffs obscure what actually changed in code by treating changes at the entity level instead. It parses code with tree-sitter to extract functions, classes, and methods as discrete entities, then diffs against those entities rather than line ranges. This means you see "function `blahh` was modified" instead of "lines x-y changed." The tool requires no setup and works in any existing Git repository. It supports 28 languages through tree-sitter and provides entity-level blame and impact analysis alongside diffs.
The tool is built specifically for coding agents and AI workflows, making it suited for teams integrating LLMs into development processes or building agent-native infrastructure. It works locally by default, with cloud-backed queries available on an opt-in basis per repository. The README emphasizes that logging in does not automatically upload repositories or send queries, with explicit consent flows, preview screens, and forget controls documented for public and private repo states.
The project maintains a substantial test suite and supports installation through multiple channels including Homebrew, npm, Bun, Cargo, and Docker. A known name conflict exists with GNU Parallel's `sem` binary, though using npm or Bun installation paths avoids this collision entirely.
