Why We Use Voyage AI for Document Embeddings

Embeddings are the backbone of semantic search in a RAG system. They convert text into numerical vectors that capture meaning, enabling the system to find relevant content even when the exact words do not match.

Rekall-IQ uses Voyage AI as the default embedding provider for document and query embeddings. Here is why.

Accuracy and Relevance

Voyage’s embedding models, particularly voyage-4-lite, deliver strong retrieval accuracy for enterprise document content. They handle the kind of formal, instructional, and policy language that appears in internal company documents.

Batched Ingestion for Stability

When a workspace admin uploads documents, Rekall-IQ processes them through a batched embedding pipeline. Text chunks are sent to Voyage in batches of 20, with a maximum of 2 concurrent requests and a 500ms delay between batches. This keeps ingestion stable even for large document sets and avoids rate-limit errors.

Provider Flexibility

Rekall-IQ is built with a provider abstraction layer. While Voyage is the default, organisations can switch to Gemini embeddings by changing environment variables. The system adapts its vector dimensions and model selection automatically.

This flexibility means Rekall-IQ can evolve as embedding technology improves, without requiring a platform rebuild.