Embedding tokens are spent primarily when you train your chatbot. This includes manual training via the Training menu and auto-retraining that occurs when a trained page is updated via the post editor or some other method. A small number of embedding tokens are also used whenever someone messages the chatbot.
Embedding tokens are dramatically cheaper than input and output tokens. For instance, you can likely train 1,000 pages for $0.01 or less. Most months, you will probably spend just a few cents on embedding tokens.
We call embedding tokens “training tokens” in the Resolver plugin.
What is embedding?
Embedding is the process of converting plain text into vectors. Resolver uses embedding APIs to send your site’s text to an embedding model, which returns vectors that are stored in your database.
When someone asks a question, Resolver also embeds it, then mathematically compares the resulting vectors to the ones stored in your database to find the most relevant content on your site to share with the chat model.
You can read more about embedding providers here if you are curious: