Repository-aware AI code review.
Independent engineering project
CodeRift retrieves dependency context before reviewing a code change.
01 / CONTEXT
Why repository context matters.
A diff may omit callers, downstream dependencies, and repository conventions. CodeRift uses program analysis to retrieve relevant context before the review stage.
02 / ARCHITECTURE
System architecture.
CodeRift builds a graph from AST-derived call relationships, retrieves relevant functions and dependencies, and gives that context to a Reviewer. An Auditor then critiques the initial findings before validated issues become the final review.
Changed files connect to related functions and dependencies through AST-derived call graphs.
03 / REVIEW PIPELINE
Reviewer and Auditor stages.
The Reviewer identifies potential issues. The Auditor checks grounding, relevance, and confidence before findings are included in the final output.
04 / IMPLEMENTATION
Current implementation.
The system combines repository-wide context, two review stages, GitHub pull-request automation, and a VS Code extension.