DeltaLens

<p align="center"> <h1 align="center">DeltaLens</h1> <p align="center"> <strong>Semantic-diff-aware code intelligence for AI coding assistants</strong> </p> <p align="center"> <a href="https://github.com/YashNamdeo/DeltaLens/actions/workflows/ci.yml"></a> <a href="https://github.com/YashNamdeo/DeltaLens/blob/main/LICENSE"></a> <a href="https://github.com/YashNamdeo/DeltaLens/stargazers"></a> </p> </p>

---

DeltaLens sits between your codebase and any AI coding assistant (Claude Code, Cursor, Windsurf). It combines Tree-sitter static analysis with diff-semantic classification to send the minimum viable context — not just fewer files, but the *right representation* of those files.

> 15-40x token reduction | < 1s incremental updates | Works with any MCP-compatible AI tool

Every time an AI assistant reviews a change, it scans the codebase. On a 500-file repo that's ~150k tokens. Most of that context is noise — files not affected by the change.

Existing tools achieve recall by flagging 5-10x more files than necessary. DeltaLens targets precision while maintaining recall.

git diff --> Change Classifier --> Impact Scorer --> Token Allocator --> AI Assistant
               (interface          (0-1 per           (full / sig /       (minimal,
                vs impl)            node)              summary)            precise)

The key insight: classify *what kind of change happened* before deciding how much context to send.

| Change Type | What Happens | Context Sent | |-------------|-------------|-------------| | **Interfac