Endure

Endure is an AI-driven platform designed to provide code maintenance intelligence by analyzing software repositories to identify files that may pose future maintenance risks. The tool addresses the challenges of codebases losing context and accumulating technical debt by automatically evaluating each file for complexity, churn, staleness, and historical risk. Through this analysis, it generates a Debt Score and an overall Health Score for the repository, helping developers and teams pinpoint areas that require attention before they become problematic.

The platform operates by allowing users to input either a public GitHub URL or a local repository path. Within approximately 60 seconds, Endure delivers a comprehensive overview that includes a health score ranging from 0 to 100, a severity breakdown, and a list of the highest debt files. Each file is ranked by actual risk and can be sorted or filtered by debt score, complexity, churn, severity, or directory. Detailed file breakdowns are available, and users can explore co-change graphs that reveal files frequently modified together, highlighting hidden dependencies and potential blast radii within the codebase.

Endure's knowledge base is built from the repository's own git history rather than generic coding rules. It learns recurring risk patterns specific to the codebase, assigns confidence scores to these patterns, and provides CLI commands for manual verification. The platform also tracks trends over time by storing each analysis as a snapshot, allowing users to monitor whether technical debt is increasing or decreasing and to correlate changes with specific events such as deployments or refactors. For high-debt files, Endure uses a large language model to infer the intended purpose of the file, storing this intent in a database for future reference and analysis.

Currently available as a research preview, Endure is a working product that runs analyses on real codebases and seeks user feedback to validate its approach. It integrates with existing workflows without requiring manual annotations or process changes.