​

Embedding Models

Embedding models convert text into vector representations, enabling semantic search and knowledge base operations in Xagent.

​

What are Embeddings?

Embeddings are numerical representations of text that capture semantic meaning. Similar concepts have similar embeddings, allowing Xagent to:

• Find relevant information in knowledge bases
• Compare document similarity
• Power semantic search
• Enable RAG (Retrieval-Augmented Generation)

​

Purpose

Knowledge Base Operations:

• Index uploaded documents
• Enable semantic search across documents
• Retrieve relevant context for tasks

How it works:

1. Documents are chunked and converted to embeddings
2. Stored in a vector database
3. When a task queries the knowledge base, Xagent searches for similar embeddings
4. Retrieved content is provided as context to the LLM

​

When to Configure

Required for:

• Using knowledge base functionality
• Uploading and searching documents
• Building RAG systems
• Semantic document retrieval

Use cases:

• Customer support (search FAQs, documentation)
• Research assistant (query paper/database)
• Legal analysis (search case law, contracts)
• Technical documentation (search manuals, guides)

​

Supported Providers

​

OpenAI & OpenAI-compatible

Models:

• text-embedding-3-small
• text-embedding-3-large
• text-embedding-ada-002

Setup:

1. Get API key from OpenAI Platform
2. For compatible services, provide base URL and API key
3. Select embedding model

Best for:

• General-purpose embeddings
• Wide availability
• Strong ecosystem

​

DashScope (Alibaba Cloud)

Models:

• text-embedding-v4
• text-embedding-v3
• text-embedding-v2

Setup:

1. Get API key from DashScope Console
2. Configure API key
3. Select embedding model

Best for:

• Chinese language optimization
• Cost-effective for Asian languages
• Alibaba Cloud integration

​

Xinference

Models:

• bge-large-en-v1.5
• bge-base-en
• All-MiniLM-L6-v2
• Other HuggingFace embedding models

Setup:

1. Deploy Xinference server
2. Launch embedding model (e.g., bge-base-en-v1.5)
3. Configure base URL in Xagent
4. Select model from Xinference

Best for:

• Data privacy
• Self-hosted deployment
• Custom model support
• Using HuggingFace open-source models

​

Configuration

​

Step 1: Add Embedding Provider

1. Go to Models in the left sidebar
2. Click Add Model or Add Provider
3. Select embedding provider

​

Step 2: Configure Model

1. Select the embedding model
2. Enter API credentials
3. Configure parameters (dimensions, etc.)
4. Test the connection

​

Step 3: Set as Default

Set as the default embedding model for:

• Knowledge base indexing
• Semantic search operations
• Vector database operations

​

Model Parameters

​

Dimensions

The size of the embedding vector. Considerations:

• Lower dimensions (384-768): Faster, less storage, good for simple tasks
• Medium dimensions (768-1536): Balanced performance
• Higher dimensions (3072+): Better accuracy, more storage/cost

Trade-offs:

• Higher dimensions = better semantic understanding
• But increased storage and computation
• Choose based on your requirements

​

Chunk Size

When documents are indexed, they’re split into chunks before embedding. Typical sizes:

• 512-1000 tokens for general documents
• Larger for technical documentation
• Smaller for precise retrieval

Impact:

• Smaller chunks: More precise retrieval, more chunks to manage
• Larger chunks: More context per chunk, less precise

​

Best Practices

​

Choose the Right Model

For general use:

• OpenAI text-embedding-3-small
• Balances cost and performance

For highest quality:

• OpenAI text-embedding-3-large
• BAAI/bge-large-en-v1.5

For cost-sensitive:

• Open-source models (HuggingFace)
• Self-hosted options

​

Optimize Chunking

• Match chunk size to typical query length
• Consider document structure (paragraphs, sections)
• Test different chunk sizes for your use case
• Use overlap to maintain context

​

Monitor Performance

• Track retrieval quality
• Monitor search latency
• Evaluate storage requirements
• Adjust based on usage patterns

​

Troubleshooting

​

Poor Search Results

Check:

• Embedding model quality
• Chunk size and strategy
• Query formulation
• Document content quality

Try:

• Using a higher-quality model
• Adjusting chunk size
• Improving document formatting
• Adding more relevant documents

​

High Latency

Optimize:

• Use smaller/faster models
• Reduce embedding dimensions
• Implement caching
• Consider vector database optimization

​

Storage Issues

Solutions:

• Use lower-dimensional embeddings
• Implement efficient chunking
• Clean up unused embeddings
• Consider storage compression

​

Integration with Knowledge Base

Embedding models work seamlessly with Xagent’s knowledge base:

1. Upload documents to knowledge base
2. Automatic indexing using embedding model
3. Semantic search powered by embeddings
4. Context retrieval during task execution

​

Next Steps

• Knowledge Base Overview - Learn about knowledge features
• LLM Models - Configure language models
• Vision LLMs - Configure vision models
• Model Overview - Understanding all model types