Weave: A Smarter Way to Merge Code Changes

Ataraxy Labs has released Weave, a new tool that helps developers merge code changes more cleanly by understanding what the code actually does, not just what the raw text looks like.

When multiple developers work on the same codebase, Git — the standard tool for managing code versions — has to combine their changes together. It does this by comparing lines of text across the versions. That works fine most of the time, but it trips up in a common scenario: two developers independently write new methods in the same class, or modify different functions in the same file. Git sees that they touched lines near each other and flags a conflict, even though the changes are completely separate from a code perspective. The tool cannot see the forest; it only sees overlapping trees.

Weave solves this by parsing the actual code structure. It uses tree-sitter, a parser that builds a logical map of how functions, classes, and methods fit together in a file. Once Weave understands the code's shape, it can tell whether two changes genuinely overlap (say, both modifying the same function) or just live nearby (adding two different methods to the same class). In the latter case, there is no conflict.

The implementation is written in Rust and integrates directly into Git's merge process — no special workflow changes needed. Developers configure it once using a standard Git file called .gitattributes, and it works from there on.

Ataraxy Labs also maintains Sem, a companion project that applies the same structural thinking to the broader task of tracking how code changes over time, not just resolving merge collisions.

The timing of this release reflects how code is being written today. AI coding assistants like Copilot and Cursor have made individual developers more productive, and they tend to generate self-contained units: a new method here, a refactored class there. That style of development — more contributors, more parallel work, more autonomously generated patches all touching the same files — creates exactly the kind of spurious conflicts that line-by-line merging handles worst.

There is a larger strategic signal worth considering here. Ataraxy Labs frames its mission as building software for agent reliability — meaning software that helps autonomous code agents and human developers work together safely. Weave and Sem read as bets on a near-term future where AI agents commit code to repositories alongside humans at meaningful scale. In that world, a merge tool that cannot distinguish between separate functions and mere text proximity stops being a minor annoyance and becomes a real bottleneck.

One practical constraint to understand before adoption: Weave relies on tree-sitter's language grammars, which vary in how well they cover different programming languages. Tree-sitter supports many languages, but not all grammars are equally mature or accurate. That means the accuracy of Weave depends directly on the quality of the grammar for your language. Any team considering Weave should first check whether tree-sitter handles all the languages your codebase uses, and at what level of correctness. This is not a flaw unique to Weave — it is a real inherited limitation of how the approach works — but it is the first question to answer.

The project is open source. For teams already managing code written or edited by AI assistants, it is worth evaluating whether Weave could eliminate some of the merge conflicts you are already seeing.