Show HN: Cross-repository understanding using static analysis and selective AI
Hacker News - AI
Jul 21, 2025 15:35
kanwarb
1 views
hackernewsaidiscussion
Summary
A new tool combines static analysis (using Tree-sitter for AST parsing) and selective AI to map dependencies and relationships across multiple code repositories, enabling real-time impact analysis when making code changes. Unlike existing tools, it relies on static analysis for accuracy and uses AI only for high-level semantic pattern matching, addressing the limitations and error rates of pure AI code analysis. This approach highlights a growing trend in the AI field toward hybrid systems that leverage AI for context and pattern recognition while relying on deterministic methods for precise technical tasks.
Built this because I kept losing context when working across multiple services. You know the problem: "what breaks if I change this auth function?" requires manually tracing through 5+ repositories to find all the dependencies. Technical approach: - Tree-sitter AST parsing for reliable dependency extraction across languages - Graph storage for efficient relationship traversal - AI only for semantic pattern matching on structured data (not raw code analysis) - Real-time impact analysis without the accuracy problems of pure semantic approaches Key insight: Don't trust AI for detailed code analysis (research shows 15-20% error rates), but use it for broader connections between components that static analysis has already verified. Differentiation from existing tools: - Sourcegraph: excellent single-repo navigation, limited cross-repo understanding - Glean: searches documentation, this analyzes actual code relationships - GitHub dependency graph: package-level tracking, this maps business l