My Architecture Lens

The problem

Stepping into a legacy codebase or a massive open-source repo usually involves weeks of manual tracing and mental mapping. Static documentation is often stale, and standard IDE navigation fails to visualize complex cross-module dependencies and architectural layers.

What it does

• Runs a 7-phase multi-agent pipeline to scan files, extract entities, and map dependencies into a JSON knowledge graph.
• Generates an interactive local dashboard with force-directed graphs, architecture layer color-coding, and fuzzy semantic search.
• Builds automated "Guided Tours" and plain-English summaries to explain code purpose to different personas.
• Performs impact analysis on git diffs to show exactly which downstream components are affected by a change.
• Maintains graph accuracy via post-commit hooks that incrementally update only modified files using WASM-based fingerprints.

Frameworks & tools

Works with any language via tree-sitter (WASM). Includes native integration for Claude Code, Node.js, and 14+ CLI platforms.

Why this beats prompting it yourself

Standard LLM prompts lack the global context of a large repo and suffer from hallucinations regarding deep dependencies. This tool combines deterministic AST parsing with semantic LLM analysis to create a commit-able, verifiable source of truth that functions without an API key once generated.

Use cases

• Accelerate developer onboarding from two weeks to twenty minutes.
• Visualize architectural impact before refactoring critical service layers.
• Search codebases semantically to find validation logic or specific domain flows.
• Share interactive architecture maps with team members without requiring LLM access.

Known limitations

Initial full scans of massive repositories incur LLM token costs for semantic extraction. Requires Node.js for the dashboard viewer.