Embedding Models

Choose an embedding provider based on your requirements:

Supported Providers

OpenAI

Cloud-based embeddings via OpenAI API

Azure OpenAI

OpenAI models via Microsoft Azure

Nebius

Multilingual embeddings optimized for European languages

Common Configuration

All providers support these parameters:

Parameter	Type	Default	Description
`provider`	string	-	Provider name
`rpm`	integer	3000	Requests per minute
`tpm`	integer	1000000	Tokens per minute

Basic Setup

embedding_model:
  provider: "openai"  # or azure, nebius
  # Provider-specific parameters...

Selection Guide

Cloud Providers (OpenAI, Azure, Nebius):

No local hardware required
Pay-per-use pricing
Easy to get started

Self-hosted OpenAI-compatible services:

Keep embeddings inside your own environment
Run the model outside the Vault process
Configure Vault with provider: "openai" and the service base_url

Important: Changing embedding models requires complete reindexing of all documents. Choose carefully before initial setup.

Comparing Embedding Quality

Test Criteria

Language: German and English document search
Domain: Domain-specific vs. general content
Length: Short vs. long text passages
Cost: API costs vs. hardware investment

Migration Between Models

Preparation

Backup: Back up current configuration
Test Environment: Test new model separately
Time Planning: Plan downtime

Migration Process

Stop DataVault
Clear Weaviate database
Activate new embedding configuration
Restart DataVault
Wait for complete reindexing

Support and Troubleshooting

Common Issues

API Limits: Adjusting rpm/tpm parameters
Authentication: Checking API keys
Performance: Check service latency and embedding batch size

Contact

For specific questions about embedding configuration: 📧 Enterprise Support: enterprise@meingpt.com