Bundle Analyzer Plugin

The bundleAnalyzerPlugin is a built-in Rolldown plugin that emits a detailed report describing your bundle's chunks, modules, dependencies, and reachability information. The report can be consumed by visualization tools, custom scripts, or LLM-based coding agents.

EXPERIMENTAL

This plugin is currently experimental and is exported from rolldown/experimental. Its API may change in future releases.

Usage

Import and use the plugin from Rolldown's experimental exports:

import { defineConfig } from 'rolldown';
import { bundleAnalyzerPlugin } from 'rolldown/experimental';

export default defineConfig({
  input: 'src/main.js',
  output: {
    dir: 'dist',
    format: 'esm',
  },
  plugins: [bundleAnalyzerPlugin()],
});

After running the build, the plugin emits an analysis file alongside your bundled output (by default dist/analyze-data.json).

Options

fileName

• Type: string
• Default: 'analyze-data.json' when format is 'json', 'analyze-data.md' when format is 'md'

The filename used for the emitted analysis asset. The file is emitted into the same output directory as the rest of the bundle.

bundleAnalyzerPlugin({
  fileName: 'bundle-analysis.json',
});

format

• Type: 'json' | 'md'
• Default: 'json'

Selects the output format.

• 'json' produces a structured data file suitable for programmatic analysis or third-party visualizers.
• 'md' produces a markdown report tailored for LLM consumption (see Markdown Format below).

bundleAnalyzerPlugin({
  format: 'md',
});

JSON Format

When format is 'json' (the default), the emitted file contains a structured object with the shape below. The timestamp field is milliseconds since the Unix epoch.

{
  "meta": {
    "bundler": "rolldown",
    "version": "1.0.0",
    "timestamp": 1705314645123,
  },
  "chunks": [
    {
      "id": "chunk-main",
      "name": "main-abc123.js",
      "size": 45230,
      "type": "static-entry", // or "dynamic-entry" or "common"
      "moduleIndices": [0, 1, 2],
      "entryModule": 0,
      "imports": [
        {
          "targetChunkIndex": 1,
          "type": "static", // or "dynamic"
        },
      ],
      "reachableModuleIndices": [0, 1, 2, 3, 4],
    },
  ],
  "modules": [
    {
      "id": "mod-0",
      "path": "src/main.js",
      "size": 3450,
      "importers": [1, 2],
    },
  ],
}

The JSON output can be uploaded to community visualizers such as chunk-visualize, or processed by custom scripts to track bundle metrics over time.

Markdown Format

When format: 'md' is set, the plugin emits a structured markdown report instead of JSON. The report is designed to be consumed by LLM-based coding agents, so you can pipe it directly into a prompt for review and refactoring suggestions.

The report is organized into the following sections:

Section	Description
Quick Summary	Total output size, input module count, entry points, and number of code-split (common) chunks.
Largest Modules by Output Contribution	All modules sorted by size, with each module's percentage share of the total output.
Entry Point Analysis	For each entry: its output filename, bundle size, the chunks it loads, and the modules it bundles.
Dependency Chains	Modules imported by multiple files, useful for understanding why a module ends up in the bundle.
Optimization Suggestions	Actionable suggestions with severity levels (see below).
Full Module Graph	Complete per-module dependency information (imports, imported-by, size).
Raw Data for Searching	Grep-friendly lines using [MODULE:], [OUTPUT_BYTES:], [IMPORT:], [IMPORTED_BY:], [ENTRY:], [CHUNK:] tags.

Optimization Suggestions

The suggestions section identifies modules that live in shared common chunks but are only reachable from a single static entry. Such modules are unnecessarily shared and could be moved closer to their entry point by enabling entriesAware: true on your output.codeSplitting groups, which is the same fix the report's own optimization tip recommends.

Each suggestion is tagged with a severity level based on the proportion of single-entry-reachable module size within the common chunk:

• [HIGH]: greater than 50%
• [MEDIUM]: between 30% and 50%
• [LOW]: less than 30%

Piping the Report into an LLM

Because the report is plain markdown, you can feed it directly to an AI assistant for review:

# After running your build
cat dist/analyze-data.md | your-cli-coding-agent "review this bundle and suggest improvements"

Example

A runnable example is available in the examples/bundle-analyzer-demo directory of the Rolldown repository. It demonstrates a multi-entry project that produces interesting optimization suggestions when analyzed with format: 'md'.