DocuMint - Code Documentation Generator for VS Code

DocuMint is a VS Code extension that generates code documentation for an entire workspace, a selected folder, or a selected file using providers such as OpenAI, Anthropic, OpenRouter, DeepSeek, or a custom OpenAI-compatible endpoint.

Generated documentation preview:

It produces:

• docs/documentation.md
• docs/documentation.html

Table of Contents

• What It Does
• What's New in 1.0.4
• How It Works
• Supported Providers
• Supported Languages
• Install
• Quick Start
• Commands
• Configuration
• Documentation Modes
• Output
• Project Structure
• Development
• Known Limitations
• Contributing
• License

What It Does

DocuMint scans source files, sends code to the configured provider, and builds project-level documentation with:

• Project overview and stats
• Per-file documentation sections
• Detected imports, exports, symbols, and project links
• Quality notes when generated docs miss detected source facts
• Markdown and/or HTML output
• HTML table of contents with navigation and search
• Mermaid diagram rendering support in generated HTML

The extension runs directly inside VS Code through a sidebar webview.

What's New in 1.0.4

DocuMint 1.0.4 is focused on speed, cache reuse, and a cleaner premium generated-document experience.

Headline: super fast generation after the first run. The first generation builds full docs and cache; the second generation can reuse unchanged file docs, project overview data, and visual blueprint cache, so repeat runs are much faster.

1.0.4 Highlights

• Super fast generation flow: file documentation runs in parallel, with the default parallel request limit set to 15.
• Smarter local CPU prep: DocuMint analyzes source structure, dependencies, symbols, imports, and prompt context locally before calling the provider.
• First run builds cache, second run is fast: unchanged docs, project overview, and visual blueprints can be reused instead of regenerated.
• Old cache wipe button: clear stale documentation and visual cache directly from the sidebar.
• Cleaner generated HTML: default dark theme, improved layout width, fixed right-side gap, and better dark-mode readability.
• Premium navigation: wider sidebar, VS Code-style project tree, folder/file icons, and + / - folder controls.
• Better project flow: HTML content follows the same order as the Project Tree.
• Visual upgrades: colored Module Scale Chart pie view, visual blueprint sections, architecture map, code workflow, and interactive dependency graph.
• Cleaner output: empty "no data found" style sections are removed when useful data is not available.
• UI Storyboard removed from generated HTML so the output stays source-grounded and avoids fake-looking UI mockups.
• Better provider support: DeepSeek and custom OpenAI-compatible endpoints are supported.
• Updated default model: new installs default to gpt-5.4-nano.
• Cleaner VSIX output: generated docs are excluded from the packaged extension.

Estimated 100-Page Website Generation Time

These are practical estimates, not fixed benchmarks. Actual time depends on provider speed, selected model, rate limits, project size, file changes, and network latency.

Design Improvements

Visual Preview

Interactive Dependency Graph:

Editable Diagram Export:

How It Works

1. Resolve provider and model from sidebar payload or VS Code settings.
2. Scan workspace files through WorkspaceScanner.
3. Optionally narrow generation to selected file/folder path.
4. Analyze source files for imports, exports, symbols, TODOs, and internal links.
5. Build a project map and file-level context from verified source facts.
6. Prepare local CPU context for each file, including dependency graph links, symbols, imports, and prompt inputs.
7. Generate detailed file documentation in parallel via the selected provider.
8. Validate generated docs against detected symbols.
9. Save output into docs/ as Markdown, HTML, or both.

Core pipeline entry point: src/services/docGenerator.ts.

Supported Providers

Configured using aiDocGenerator.aiProvider:

• openai
• anthropic
• openrouter
• deepseek
• custom

Provider implementations live in src/providers/.

Supported Languages

Current workspace scanner settings can include:

• TypeScript (.ts, .tsx)
• JavaScript (.js, .jsx)
• Python (.py)
• Java (.java)
• C / C++ / C# (.c, .cpp, .cs)
• Go (.go)
• Rust (.rs)
• PHP (.php)
• Ruby (.rb)
• Swift (.swift)
• Kotlin (.kt)
• Scala (.scala)
• Shell, YAML, JSON, XML, HTML, CSS, SCSS, and SQL

Notes:

• aiDocGenerator.targetLanguages controls which languages are scanned.
• Exclusions include node_modules, dist, build, .git, docs, test/spec files, lockfiles, .env files, logs, and other common non-source artifacts.

Install

Marketplace

Install from VS Code Marketplace:

• https://marketplace.visualstudio.com/items?itemName=wonderertech.documint

Or run:

ext install wonderertech.documint

VSIX

code --install-extension documint-1.0.4.vsix

Build from Source

git clone https://github.com/Wonderer-Tech/documint.git
cd documint
npm install
npm run compile

Quick Start

1. Open a project folder in VS Code.
2. Open the Documint view in the activity bar.
3. Select provider and model.
4. Run Configure API Key if using a cloud provider.
5. Click Generate Documentation for workspace scope, or use File / Folder quick buttons.
6. Open generated files from docs/.

Commands

Contributed commands:

• aiDocGenerator.generateDocumentation - Generate documentation
• aiDocGenerator.cancelGeneration - Cancel generation
• aiDocGenerator.configureApiKey - Configure API key
• aiDocGenerator.clearCache - Clear old documentation and visual cache

Internal scope commands used by sidebar:

• aiDocGenerator.pickAndGenerateFile
• aiDocGenerator.pickAndGenerateFolder

Configuration

All settings are under aiDocGenerator.

Key Settings

• aiDocGenerator.aiProvider (openai by default)
• aiDocGenerator.model (gpt-5.4-nano by default)
• aiDocGenerator.documentationDepth (simple | basic | standard | comprehensive)
• aiDocGenerator.outputFormat (markdown | html | both)
• aiDocGenerator.targetLanguages (language hint list)
• aiDocGenerator.maxTokens
• aiDocGenerator.temperature
• aiDocGenerator.rateLimitDelay
• aiDocGenerator.concurrentRequests
• aiDocGenerator.excludePatterns
• aiDocGenerator.customApiEndpoint
• aiDocGenerator.generateUmlDiagrams
• aiDocGenerator.enableDiffTracking

Example settings.json

{
  "aiDocGenerator.aiProvider": "openai",
  "aiDocGenerator.model": "gpt-5.4-nano",
  "aiDocGenerator.documentationDepth": "standard",
  "aiDocGenerator.outputFormat": "both",
  "aiDocGenerator.maxTokens": 4000,
  "aiDocGenerator.temperature": 0.3,
  "aiDocGenerator.concurrentRequests": 15,
  "aiDocGenerator.rateLimitDelay": 1000,
  "aiDocGenerator.excludePatterns": [
    "**/node_modules/**",
    "**/dist/**",
    "**/build/**",
    "**/.git/**"
  ],
  "aiDocGenerator.generateUmlDiagrams": true,
  "aiDocGenerator.enableDiffTracking": true
}

DeepSeek example:

{
  "aiDocGenerator.aiProvider": "deepseek",
  "aiDocGenerator.model": "deepseek-v4-flash"
}

Custom endpoint example:

{
  "aiDocGenerator.aiProvider": "custom",
  "aiDocGenerator.model": "your-model-name",
  "aiDocGenerator.customApiEndpoint": "https://api.example.com/v1/chat/completions"
}

Documentation Modes

Use aiDocGenerator.documentationDepth to control how detailed the generated documentation should be.

Notes:

• Project tree, architecture visuals, workflow diagrams, whiteboard-style diagrams, and interactive dependency graphs are available in generated HTML output across all modes.
• simple, basic, and standard can batch small files for faster generation.
• comprehensive can batch small files, but large files are not truncated. They are generated as full single-file requests, and files that exceed the provider context window are split into chunks before the output is merged.

Output

Generated output is written to a docs/ directory in the workspace root:

• documentation.md
• documentation.html

The HTML renderer includes:

• Sidebar TOC
• Route-based folder/file navigation
• Section anchors
• Search
• Theme toggle
• Syntax highlighting
• Mermaid rendering
• Project tree
• Architecture blueprint
• Draw.io export for Mermaid diagrams
• D2 source export
• Excalidraw-style whiteboard diagram
• Interactive dependency graph
• Copy-to-clipboard for code blocks

Project Structure

src/
|-- analyzer/
|   `-- sourceAnalyzer.ts         # Static import/export/symbol discovery
|-- extension.ts                  # Activation and command wiring
|-- types.ts                      # Shared types and error models
|-- config/
|   |-- secretStorage.ts          # VS Code secret storage wrapper
|   `-- apiKeyConfiguration.ts    # API key configuration helper
|-- scanner/
|   `-- workspaceScanner.ts       # Workspace file discovery and filtering
|-- providers/
|   |-- aiProvider.ts             # Base provider and prompt/chunking pipeline
|   |-- providerFactory.ts        # Provider resolution and creation
|   |-- openaiProvider.ts
|   |-- anthropicProvider.ts
|   |-- openrouterProvider.ts
|   |-- deepseekProvider.ts
|   `-- customProvider.ts
|-- services/
|   |-- docGenerator.ts           # Orchestration + writing docs output
|   |-- documentationValidator.ts # Checks generated docs against source facts
|   |-- htmlTemplate.ts           # Full HTML document template
|   `-- modelMetadataService.ts   # Context window metadata fetch/cache
`-- views/
    `-- sidebarProvider.ts        # Sidebar UI webview and state sync

Development

Prerequisites

• Node.js 18+
• npm 9+
• VS Code 1.110.0+

Commands

npm install
npm run compile
npm run typecheck
npm run watch
npm test

Package extension:

npx @vscode/vsce package

Known Limitations

• Static analysis is intentionally lightweight. It improves accuracy, but it is not a full compiler for every language.
• Documentation quality still depends on the selected model and the source code that is available in the workspace.
• Cloud providers receive selected source code after confirmation. Use a localhost custom endpoint when code must stay local.
• aiDocGenerator.concurrentRequests defaults to 15 for faster generation. Lower it in settings if your cloud provider rate-limits requests.
• Very large demo media can make the VSIX larger than the extension code itself.

Contributing

Issues and PRs are welcome.

• Repository: https://github.com/Wonderer-Tech/documint
• Issues: https://github.com/Wonderer-Tech/documint/issues

License

MIT. See LICENSE.