Inside the System (1): From 3000 Lines to 300, Refactoring the Plugin Core
How the codebase split a monolithic main.ts into a modular architecture — and added a streaming query interface on top without bloating the runtime.
The Challenge
When we first released v1.0.0, the entire plugin lived in a single main.ts file of nearly 3,000 lines. It worked — multi-page generation, structured JSON prompts, bidirectional links, and the Lint system. But it was a monolith. Adding new features meant navigating thousands of lines of intertwined logic.
The Refactor: 9 Focused Modules
In v1.2.0, we extracted the codebase into 9 focused modules. The result: main.ts shrank to just 305 lines — a 90% reduction.
- Ingestion Engine — Extracts entities and concepts from source notes
- Query Engine — Streaming ChatGPT-style dialog with Markdown rendering
- Lint System — Detects duplicates, dead links, orphans, empty pages
- Auto-Maintenance — File watcher, periodic lint, startup health check
- Settings UI — Step-by-step configuration flow
- Page Factory — Generates entity, concept, and summary pages
- LLM Clients — Unified interface for Anthropic, OpenAI, Gemini, and more
Why this matters: Each module has a single responsibility, clear interfaces, and can be tested independently. The rapid iteration pace of v1.7.x was only possible because the foundation was solid.
The Query Interface
The Conversational Query Interface lets you ask questions about your own knowledge:
- Streaming responses with Markdown rendering
- Multi-turn conversations with history
- Wiki-links embedded in answers as breadcrumbs
- Save to Wiki — valuable conversations become Wiki pages
This transformed the plugin from a knowledge organizer into a knowledge partner.