Getting Started

Getting Started

This page walks the core loop PolyGraph is built around. If you haven't installed it yet, see Installation.

1. Feed it a project

There are two ways to get code into PolyGraph:

• Scan a path (default). Paste an absolute folder path into Scan a folder. The analysis sidecar walks that directory on disk, skipping node_modules, build output, .git, and very large files, into a { path → source } map. Nothing leaves your machine.
• In-browser picker (fallback). Choose a folder; the browser reads its files into the same map and sends it to the sidecar's /analyze endpoint. Handy in the web version when you don't want to type a path.

Large repositories. Above ~3000 files the scan pauses and asks you to confirm before doing the full analysis (a guard against accidentally pointing it at something enormous). Confirm to proceed. See Troubleshooting & FAQ.

Once analysis finishes you get an interactive graph: each file is a card (collapsed by default), with edges for the relationships between files.

2. The core loop

This five-step loop is the heart of the product:

1. Select a node. Click a card. The detail panel shows its incoming/outgoing edges plus metadata (role, environment, runtime). Expand a file to reveal its members (classes, functions, methods…).
2. See its connections. Selecting a card highlights its direct neighbors; the rest dims. Shift-click a second card to highlight the shortest path between the two (or tell you there's no connection).
3. Blast radius. Ask what a change to this node would affect — everything transitively downstream, grouped by package / file / kind. See Impact Analysis & Insights.
4. Edge evidence. Click an edge to see exactly where the relationship was observed — file · line · column · provider — and the underlying symbol→symbol relationships behind an aggregated file edge. See The Graph Model.
5. Open in editor. From a node's detail panel, preview the source inline (syntax-highlighted) and jump to the exact line in VS Code or a JetBrains IDE (desktop only). See Exports, Workspaces & Editor.

3. Shape the view

• Layouts & Navigation — switch layout engines, filter by kind/language/folder, search, and enter focus mode.
• Query Language — isolate subgraphs with a small text query (incoming:>5, role:react-component, depends-on:"database", …).
• Abstraction levels — zoom out from files to package and workspace level using the project's manifests. See The Graph Model.

4. Get it out

• Export the current graph to DOT, GraphML, Mermaid, JSON, SVG, or a standalone HTML report.
• Save a workspace to capture the current filters, layout, camera, and pins to restore later.

Running the web / CLI version

End-users don't need this, but if you're working from source:

bun install
bun run dev      # Next.js dev server → http://localhost:3003 · sidecar on a loopback port

Open http://localhost:3003 (needs a WebGPU-capable browser), paste an absolute folder path, and explore. The same kernel powers the architecture-rules CLI — see CLI: Rules & Diff and Development & Building.