A new command-line tool, Slopo, tackles a persistent challenge in software development: detecting non-exact code duplication. Unlike tools that catch simple copy-pastes or nearby similar blocks, Slopo uses embedding models to identify code snippets that perform similar functions but are written differently and scattered across a codebase. This approach aims to uncover the most insidious forms of duplication that often escape human review and traditional static analysis.

Slopo operates by calculating embeddings for code units and then identifying pairs with close vector representations. While similarity doesn't equal duplication, these pairs become candidates for review. The tool supports Python, JavaScript, Java, Kotlin, C#, and Go. It generates reports that cluster similar code units, ranked by similarity and codebase distance, providing input for AI coding agents to confirm actual duplication.

Installation is straightforward via `uv`, and setup involves a simple `slopo init` command. Users configure the tool with a source directory and an embedding model, with options for providers like Voyage AI. The analysis workflow includes incremental indexing and an ignore file, allowing developers to manage reviewed clusters and refine refactoring efforts. The tool's output helps identify areas needing refactoring, especially within parser code, as demonstrated in its own example report.

This offers a more nuanced approach to code quality, moving beyond simple textual matches to semantic understanding. By leveraging AI-powered embeddings, Slopo helps developers surface and address subtle code redundancies that can hinder maintainability and introduce bugs. The tool's integration with AI coding agents streamlines the refactoring process by automating the confirmation of duplication and management of reviewed code.