31 out of 31 merge scenarios across 7 languages, 100% accuracy, compared to Git’s 48% and mergiraf’s 83%. That’s the scorecard for Weave, a semantic merge driver from Ataraxy Labs that parses code by language structure instead of lines.
Why Line-Level Merge Is Broken
Git sees overlapping lines and screams CONFLICT even when two agents edit different functions in the same file. That’s because Git has no concept of a function, a class, or a method. Weave does. It uses tree-sitter to parse source code into an AST, then merges at the entity level. Two agents touch different functions? Clean merge, every time. No conflict noise.
Weave’s Three-Layer Architecture
Weave ships three layers. First, a merge driver that plugs into Git’s ort strategy and replaces the default line-level diff. Second, a CRDT-based coordination layer where agents claim entities before editing, catching conflicts before they happen. Third, an MCP server exposing 15 tools via the Model Context Protocol, letting Claude and other AI agents call Weave directly. You can use just the merge driver or add coordination for multi-agent workflows.
31/31 on the Benchmark, Zero Regressions in Production
Ataraxy Labs tested Weave on 31 hand-crafted merge scenarios covering 7 languages. Full pass. Then they ran it on 4,917 real-world file merges across C, Python, and Go — 83 real-world wins with zero regressions. Over 1,500 downloads so far. It supports 28 languages (TypeScript, Python, Rust, Go, Java, C++, Ruby, Swift, Kotlin, Elixir, Bash, and more) plus 5 data formats like JSON, YAML, TOML, CSV, Markdown.
Install is a single brew install weave, then weave setup in your repo. After that, git merge uses Weave automatically. For teams running AI coding agents simultaneously, the CRDT layer turns merge conflicts into a solved problem rather than a daily annoyance.
Source: Weave: Merging based on language structure and not lines
Domain: ataraxy-labs.github.io
Comments load interactively on the live page.