Skip to main content
Raily works with the major model providers. They are used in three places:
  • Agentic layer: analyzes a query and rewrites it for better retrieval. See Agentic Layer.
  • Answers: generates the response from the retrieved passages in chat.
  • Embeddings: turns your content and queries into vectors. You choose the embedding provider per collection when you connect a vector store.
By default, Raily runs these model calls on its own account. To bill query analysis, chat answers, source setup, quality evaluators, and LLM reranking to your own Google account instead, add your Google AI key. See Use your own Google AI key. Embeddings and document processing still run on Raily.

OpenAI

GPT for answers and OpenAI embeddings

Anthropic

Claude for answers

Google

Gemini for the agentic layer, answers, and embeddings

Vertex AI

Google Vertex embeddings

AWS Bedrock

Bedrock embeddings

Sentence Transformers

Self-hosted open-source embeddings