Spaghetti code kills projects. Bloated context kills AI.
PanzaScope fixes both.
Stop feeding raw scripts to your LLMs. PanzaScope visualizes complex dependencies in milliseconds and extracts logic-stripped JSON blueprints, saving you hours of refactoring and up to 60% in token waste depending on the context you choose to copy.
I welcome contributions from developers, researchers, and enthusiasts across all levels of expertise.
Looking for a place to start?
Whether you have a bug fix, a new feature, or just want to share your thoughts, check the Issues tab for open tasks or open a new one to propose a feature. Look for tasks labeled with good first issue or help wanted—they are specifically designed to be highly modular and easy to pick up!
Generates two complementary outputs from a single scan:
- an interactive, hardware-accelerated HTML dependency map (Planetarium) for human exploration;
- a lightweight
_LLM_Payload.jsonblueprint optimized for AI assistants.
This separates visual analysis from machine-readable context, allowing both humans and LLMs to work with the same project efficiently.
| Planetarium: Full Project Cosmology | System Focus: Node Inspection |
 |
 |
PanzaScope doesn't just read text; it understands syntax. Using dedicated Language Processors, it surgically extracts structural blueprints (class signatures, HTML/DOM anchors, properties) while stripping away heavy logic bodies.
- Natively Supported: C#, C++, C, Python, JavaScript, Lua, Ruby, HTML, CSS, UXML, USS, JSON, XML, and YAML.
- Infinitely Extensible: Built on a modular plugin system. Need to map a new language like Rust, Go, or TypeScript? Just drop a custom Python script into the
processors/folder—the engine will automatically load it and apply your specific extraction rules. Check out our "Good First Issues" to help us expand PanzaScope's language support!
Reduces context window usage by removing implementation details while preserving project structure.
The generated blueprint contains:
- files and dependencies
- class hierarchies
- methods and signatures
- HTML DOM structure (IDs, Classes, Tags)
- architectural relationships
This provides LLMs with the information needed to understand the codebase while significantly reducing token usage.
Notice how switching from Full Code to API / Method Map Only drops the token count from ~15,600 to ~3,000, shifting the AI hallucination risk from MEDIUM to LOW.
| 🔴 Traditional: Full Code Payload | 🟢 PanzaScope: Eco-Scan Blueprint |
 |
 |
Built to work natively with Ollama and models such as qwen2.5-coder:7b.
Analyze proprietary or commercial codebases entirely offline:
- zero cloud uploads
- zero API costs
- zero data leaks
- low latency
Continuously evaluates dependency density and architectural health.
Projects are classified into three risk levels:
🟢 Stable — low coupling and modular architecture
🟡 Fragile — increasing dependency propagation
🔴 Critical — high coupling with elevated regression risk
Highlights classes or components with an unusually high number of direct dependencies (≥7), helping identify potential architectural bottlenecks before they become maintenance problems.
Automatically ignores generated reports and output folders, preventing recursive scans and infinite processing loops.
Uses file metadata to skip unchanged files, dramatically reducing scan time while enabling instant browser hot-reloading during development.
The Planetarium map uses custom orbital geometries and distinct colors to represent project taxonomic layers:
- 🟣 L0 (Contracts): Decoupled abstractions, interfaces, and API contracts.
- 🔵 L1 (Data/DNA): Serialized configurations, JSON nodes, ScriptableObjects, and style sheets (USS/CSS).
- 🟠 L2 (Core Systems): Global orchestrators, central managers, and network controllers.
- 🟢 L3 (Logic): Distinct operational scripts, gameplay loops, and business logic.
- 🔴 L4 (UI Layouts): Visual panels, Canvas components, and layout hierarchies.
- Python 3.10+
- requests library
- Ollama (Highly recommended for local offline AI assistance)
Clone the repository
git clone https://github.com/Panzadabira/PanzaScope.git
cd PanzaScopeInstall dependencies
pip install -r requirements.txtLaunch the model (Optional for local AI analysis)
ollama pull qwen2.5-coder:7b
ollama serveRun the tool
python main.pyRunning the command above will launch the PanzaScope UI Hub. You don't need to memorize any complex CLI arguments. Simply select your local project folder, choose your AI provider, and click Run. The engine will automatically generate the JSON blueprint and open the interactive Planetarium directly in your default browser.
This repository serves as the core open-source engine of PanzaScope. If you are developing complex gamedev pipelines and want a frictionless, unified workflow:
PanzaScope for Unity Editor Available for 42.00 CHF — the ultimate answer to Life, the Universe, and your spaghetti code architecture.
The Premium Unity Asset integrates the Planetarium graph directly inside Unity as a native Editor Window (using UI Toolkit), featuring automatic hot-reloads on script saves, custom editor layouts, and instant one-click pinging of files in the project Inspector.
The core engine remains free and open-source. If you find PanzaScope valuable, consider supporting its long-term development by purchasing the Premium Unity Editor Integration. https://assetstore.unity.com/packages/tools/utilities/panzascope-ai-dependency-mapper-380466
⚖️ License: Distributed under the MIT License.
Built with passion for developers by Panza Labs.