How bbawj/obsidian-semantic-search works
Project Type & Purpose
• Type: A full-stack Obsidian plugin combining a TypeScript “frontend” (Obsidian UI integration) with a Rust/WASM “core” (text processing, CSV generation, embedding).
• Purpose: Let Obsidian users generate semantic-searchable embeddings of their notes and query via a modal or inline link suggestions.
File Structure
• Top level:
• main.ts + manifest.json → entry point & Obsidian plugin definition
• esbuild.config.mjs, tsconfig.json → build pipeline for TypeScript
• release.sh, version-bump.mjs → CI/release scripts
• Cargo.toml, wasm-pack integration via esbuild → Rust/WASM module
• src/ (Rust)
• embedding.rs, file_processor.rs, generate_input.rs, obsidian.rs, error.rs → core text processing & API calls
• settings/ → Rust types for plugin settings
• src/ui (TypeScript)
• generateEmbeddingsModal.ts, queryModal.ts → UI command entrypoints
• linkSuggest.ts, suggestion.ts → context menu + suggestion component
• types/obsidian → type definitions for Obsidian APIs
• assets: styles.css
Extract Key Technical Insights from README
• Commands: Generate Input → input.csv; Generate Embedding → embedding.csv via external API; Open Query Modal → semantic search UI; Link suggestion using {{}}.
• Configurable via Obsidian “data.json” (API URL, key, model, regex, etc.)
• Dependencies: Obsidian plugin API, TypeScript, Rust/WASM, external embedding API (OpenAI, Ollama, etc.)
Main Components & Interactions
a. Components:
1. Obsidian Host – Provides plugin lifecycle, UI modals, vault FS access
2. Plugin Frontend (TypeScript) – main.ts + UI modules
3. WASM Core (Rust) – text-to-CSV, batching logic, fetching embeddings (or providing data to TS for fetch)
4. External Embedding Service – HTTP API (e.g. OpenAI)
5. File System (Vault) – input.csv, embedding.csv, plugin settings
b. Data Flow & Relationships:
1. User triggers “Generate Input” → Frontend calls WASM generate_input → WASM processes markdown, writes input.csv to vault
2. User triggers “Generate Embedding” → Frontend reads input.csv, calls WASM/embed or TS wraps fetch calls → External API returns embeddings → results written to embedding.csv
3. User triggers “Open Query Modal” or types {{}} → Frontend loads embedding.csv, prompts user, computes similarity (in WASM or TS), renders suggestions via UI modules → optionally writes links back to markdown
c. Architectural Patterns & Principles:
• Separation of concerns: UI vs. core logic
• WASM for CPU-intensive tasks and to share logic across platforms
• Clear command-pattern mapping from Obsidian commands → UI modules → core services
• Batch processing for API rate limits
d. Key Technologies/Frameworks:
• Obsidian Plugin API (host)
• TypeScript + esbuild
• wasm-pack, Rust → WebAssembly module
• Fetch/HTTP for embedding API
• CSV I/O via vault FS
Final Recommendations
• Be as detailed as possible: include settings flows (where user config lives), error handling path (Rust error.rs), and UI modules for each commands/view.
• Technology stacks next to each component.