OpenAI and Google both offer embedding APIs, which Resolver uses to convert your site’s content into vectors. These vectors store mathematical relationships between words. This lets Resolver select the most relevant content as context for the chatbot.
When you train your chatbot, it sends your content to the embedding model to convert it into a set of vectors, which it then stores in the WordPress database.
When someone asks the chatbot a question, it sends their message to the embedding model to convert it into vectors as well. Then, Resolver uses a mathematical comparison between the message vectors and the vectors from your database to find the most contextually relevant content. Finally, it sends the question and selected text to the AI chat model, which uses this context to provide an intelligent reply.
The AI chat model is never sent vectors – only plain text. That’s why it’s possible to use an embedding provider from a different company than the one you use for your chat model, e.g., using Claude for your chatbot while using Google for your embeddings.