Generate LLM-optimized dependency analysis bundles from your codebase.

pviz-parser analyzes your project's dependency graph and produces a structured JSON bundle designed to fit inside an LLM context window. Instead of pasting files one by one, you give your LLM a complete picture of how your codebase is wired — nodes, edges, import relationships, cycle detection, and per-file metrics — in a single compressed artifact.

Installation

Quickstart

pviz . -o bundle.json --clean

That's it. Point pviz at your project root, specify an output path, and get a bundle ready to drop into Claude, ChatGPT, or any LLM that accepts file uploads or large context.

Usage

pviz <scan_root> -o <output.json> [options]

Arguments:
  scan_root             Root directory of the project to analyze

Options:
  -o, --output          Output path for the bundle JSON (required)
  --clean               Nuke the artifact cache before running (recommended when switching repos)
  --store-root PATH     Override the sandbox directory (default: per-user .pviz_store)
  --mode classic|zones  Build mode (default: zones)
  --max-bytes N         Per-file size limit in bytes (default: 100MB)
  --allow-output-in-repo  Allow writing the bundle inside scan_root

Examples

# Analyze current directory
pviz . -o bundle.json --clean

# Analyze a specific project
pviz ~/projects/myapp -o ~/Desktop/myapp_bundle.json --clean

# Use a custom store root to keep artifacts organized
pviz . -o bundle.json --store-root /tmp/pviz --clean

What's in the bundle

The output is a structured JSON artifact with:

• Nodes — one per source file, with LOC, SLOC, import/exporter counts, SCC membership, symbols, and language
• Edges — directed import relationships between files
• Dependency metrics — which files are most imported, which import the most, hotspots
• Cycle detection — strongly connected components (SCCs) flagged at the node level
• Discovery manifest — full file inventory with language breakdown
• Folder index — per-file import surface and resolution data
• Summary — counts, parse status, edge stats, crosstalk candidates

A compressed format is also generated alongside the standard bundle (.compressed.json), typically 55–65% smaller, optimized for tight context windows.

Language support

Kotlin, Go, and Rust analysis requires compiled binary dependencies that are part of the hosted SaaS only. Polyglot repos with multiple supported languages are handled automatically — the bundle merges all detected languages into a single artifact.

CLI vs SaaS

pviz-parser is the open source CLI. It runs locally, produces bundles you own, and supports Python, TypeScript, JavaScript, and partial resolution for Java out of the box.

pvizgenerator.com is the hosted SaaS layer. It adds full Kotlin, Go, and Rust support, hosted bundle storage, bundle diffing across commits, and MCP delivery for direct LLM tool integration — without running anything locally.

Requirements

• Python 3.10+
• No other system dependencies for Python/TS/JS/partial-Java analysis

License

MIT — see LICENSE

Built by Michael McClellan