TinyEmbed lets you create text embeddings, perform semantic searches, and analyze text similarity without any backend server. All processing happens locally, right within your browser.

• Single Text Embedding: Generate embeddings for individual texts
• Batch Embedding: Process multiple texts at once
• Semantic Search: Create a corpus of texts and search by meaning
• Similarity Analysis: Compare two texts and analyze their semantic similarity
• Local Processing: All computation runs in your browser
• No API Keys: No external API services or keys required

• Next.js with App Router and TypeScript
• TinyLM for embedding generation
• shadcn/ui for components
• Tailwind CSS for styling
• WebGPU (optional) for hardware acceleration

• Modern browser with WebGPU support for full performance (Chrome 113+, Edge 113+, or other Chromium browsers)
• Falls back to WebGL when WebGPU is unavailable

# Clone the repository
git clone https://github.com/wizenheimer/tinyembed.git
cd tinyembed

# Install dependencies
npm install

# Start the development server
npm run dev

# Build for production
npm run build

# Start the production server
npm start

• Select the embedding model from the dropdown (currently only Nomic Embed is supported)
• Choose the encoding format (Float for raw vectors or Base64 for compact storage)
• Click "Load Model" and wait for the loading process to complete
• The status indicators will show you the current state of the model and system

• Enter text in the input field
• Click "Generate Embedding"
• View the resulting embedding vector and metadata

• Enter multiple texts, one per line
• Click "Generate Batch Embeddings"
• View the results for the entire batch

• Add documents one by one to create a searchable corpus
• Enter a search query
• Click "Search" to find semantically similar documents
• Results will be ranked by similarity

• Enter two texts to compare
• Click "Compare Texts"
• View the similarity score and interpretation

• WebGPU acceleration provides the best performance, especially for larger embedding models
• Loading models can take some time depending on your internet connection (models are downloaded the first time)
• Processing times improve significantly after the initial runs
• For using vector search over HNSW index, try out tinkerbird library.

This project uses:

• TinyLM for browser-based ML inference
• Nomic Embed text embedding model
• shadcn/ui for UI components