This appears to be a business decision by Anthropic. They are focusing mainly on text generation models, whereas OpenAI has more extensive coverage of the AI stack, including embedding models, real-time voice, image generation, etc.
If you’re using a Claude model for your chatbot, you can use Google or OpenAI as your embedding provider with full interoperability. The embedding provider does not need to match the AI model you select.
How can a different embedding provider work with Claude?
The processes are totally separate. The embedding provider converts your content and the visitor’s message into vectors. When someone messages your chatbot, their message is sent to the embedding provider, and the resulting vectors are then compared with the vectors stored for your content. This is how the most relevant content is found for their question. Lastly, their message and the most relevant content are sent to the chat model, which formulates its response. In other words, all the chat model ever sees is plain text, so it doesn’t matter which embedding model is used.