Semant is a blazingly fast, fully-local semantic code search extension for Visual Studio Code. Instead of hunting for exact keyword matches, Semant lets you search your codebase using meaning and concepts.
Because Semant uses the ONNX Runtime and Transformers.js, the entire AI pipeline runs directly inside your IDE. Your code never leaves your machine.
🌟 Features
Semantic Search: Search by what the code does, not just the variable names. (e.g., searching for "read file to string" will find filesystem buffering code even if those keywords aren't present).
100% Local AI: Powered by HuggingFace's all-MiniLM-L6-v2 embedding model. No API keys, no subscriptions, no cloud telemetry.
Dedicated Sidebar: Semant integrates beautifully into the VS Code Activity Bar with a custom tree-view interface.
Smart Caching: Semant tracks your file modifications (mtime) and only runs the heavy AI embedding models on files that have changed since your last index, making subsequent re-indexes almost instantaneous.
🚀 Getting Started
Open the Semantic Search panel in your Activity Bar (the glowing 'S' icon).
Click Index Workspace (or the refresh icon). Semant will discover, chunk, and embed your codebase. The first run takes a moment as it downloads the model locally and processes your files.
Once the index is complete, click the Search icon (🔍) in the panel header.
Type your query! The results will instantly appear in the sidebar. Click any result to jump exactly to that chunk of code.
⚙️ Configuration
You can customize Semant through your VS Code settings.json:
semant.excludePatterns: An array of glob patterns to ignore during indexing. By default, it ignores folder like node_modules, .git, dist, and binary artifacts.
semant.model: The HuggingFace model ID to use for embeddings (Default: Xenova/all-MiniLM-L6-v2).
🛠️ Building from Source
If you want to tweak Semant or run it locally:
git clone https://github.com/ogb-daniel/semant
cd semant
npm install
npm run compile
Press F5 in VS Code to open an Extension Development Host.
🔒 Privacy
Privacy is a core pillar of Semant. Your source code is run through an embedded .onnx model locally on your CPU via WebAssembly. Neither your search queries nor your code are ever transmitted across the internet.
