shipped embeddings on the ngrok AI gateway today. favorite detail:
send a fallback list with two different embedding models and we 400 you.
vectors from different models don't compare. silently failing over to another one would quietly wreck your index. so failover only goes to another provider of the same model.
heads up if you run the ngrok agent behind a corporate TLS proxy: check your config for connect_cas: host.
this is a third-party CA issue (emudhra), and could impact your ngrok config. default config users aren't affected. details: https://t.co/omn3q9AEoD