Skip to content

IntentumDiff

Semantic code review — see what a change means, not which lines moved.

A textual diff shows you that 40 lines changed. It cannot tell you whether those 40 lines are a rename, a function moved between files, a reformat, or a subtle change in behaviour hiding among them. Reviewers spend most of their attention separating those cases by hand.

IntentumDiff parses both sides into a syntax tree, matches the trees against each other, and classifies every change by what it does:

Category Meaning
Meaningful Behaviour changed — this is what review time is for
Refactoring Structure changed, behaviour did not
Moved The same code, somewhere else
Ignored style Formatting and whitespace

A 400-line reformat with one behavioural change inside it stops being a wall of red and green, and becomes one change worth reading.

Start here

What it supports

78 languages ship in the box, each an independent WebAssembly parser — Python, TypeScript, Go, Rust, Java, C#, C++, SQL dialects, Terraform, and more. See the full list.

Language support is a plugin, not a feature of the engine — anyone can add a language without changing the core, and contributions are welcome.

Beyond source code it also diffs structured data (JSON, YAML, TOML, XML), documentation (Markdown, AsciiDoc, LaTeX) and images, using a perceptual comparison rather than a byte one.

Privacy

IntentumDiff runs entirely on your machine. Parsing, matching and classification are local, and nothing is sent anywhere by default.

The optional LLM explainer is bring-your-own-key and opt-in. When enabled it sends a privacy-safe fact sheet — counts, categories and flags — never your source code, identifiers or literals. See Privacy for exactly what leaves your machine.

Status

IntentumDiff is early software and currently ships as a prerelease. Please report anything that does not behave as documented — including the docs themselves.