Proxy embedding requests through the Hive model gateway

Hive’s model gateway now relays embedding requests through the new endpoint POST /inference/v1/embeddings, so OpenAI-compatible embedding clients can point at Hive instead of an upstream provider. Operators create an embedding profile, send requests using the Hive profile token as a bearer credential, and use a stable profile model name that Hive rewrites to the configured upstream embedding model before forwarding. Provider credentials stay on the server, requests are attributed to tokens, and embedding usage is recorded with an explicit embedding operation and an estimated cost from input tokens (output tokens recorded as zero).