The CLI’s teach LLM is configured separately from the
server’s embedding provider, and there’s no mechanical link between
them. Changing one doesn’t affect the other.
In practice, when you switch the server’s embedding provider, you
often want to swap teach’s extraction LLM at the same time
(same vendor, same key, same billing). But the choice is yours:
you can run OpenAI embeddings on the server and a self-hosted
Llama for teach extraction if you prefer.
Switching teach’s LLM does not require a re-embed.
teach only produces text; the server embeds the text
exactly the same way regardless of where the text came from.
See Embedding Migration for when a re-embed is needed.