Building Repository-Level Code Intelligence with Repowise and Graph Analysis

How to Build Repository-Level Code Intelligence with Repowise Using Graph Analysis, Dead-Code Detection, Decisions, and AI Context

Repowise provides a unified pipeline for repository indexing, graph analysis, and LLM-ready context generation. It automates the detection of code influence using PageRank and identifies redundant code with a configurable 0.7 safety threshold.

Why This Matters

Traditional code analysis often lacks cross-file context, making it difficult for LLMs to navigate large repositories without hallucinating dependencies. Repowise bridges this gap by creating a persistent graph-based index that captures structural influence and architectural decisions, moving beyond simple text-search to a multi-layered intelligence model.

Key Insights

• Graph-based node importance: Repowise uses PageRank to identify the most influential files in a codebase, such as signer.py in the itsdangerous project.
• Automated dead-code detection: The tool evaluates modules for deletion safety using a threshold-based approach with a default 0.7 score.
• Architectural Decision Records (ADR): Users can insert # DECISION: tags directly into source code to maintain persistent architectural context for AI tools.
• MCP-style tool integration: Provides CLI access to functions like get_dead_code, search_codebase, and get_architecture_diagram.
• Multi-LLM Support: Integrates with Anthropic’s Claude 3.5/4.5 and OpenAI’s GPT-4o models for semantic querying and codebase overview.

Working Examples

Repowise configuration file (.repowise/config.yaml) for LLM and Git intelligence setup.

provider: anthropic\nmodel: claude-sonnet-4-5\nembedding_model: voyage-3\nreasoning: auto\ngit:\n  co_change_commit_limit: 200\n  blame_enabled: true\ndead_code:\n  enabled: true\n  safe_to_delete_threshold: 0.7

Practical Applications

• Onboarding: Use PageRank visualization to identify core logic modules in legacy repositories.
• Safe Refactoring: Use dead-code detection with safe-only flags to prune unused functions without breaking dependencies.
• AI Context Enrichment: Generate CLAUDE.md to provide LLMs with updated project-specific structural metadata.

References:

• https://www.marktechpost.com/2026/05/15/how-to-build-repository-level-code-intelligence-with-repowise-using-graph-analysis-dead-code-detection-decisions-and-ai-context/