Inference overview
The three model surfaces on the platform - chat models for reasoning, embeddings for search, reranking for precision - all OpenAI-compatible.
Chat models
How a workload reaches a chat model, which environment the platform injects, how to pin a different model, and how to bring your own provider.
Embeddings
Turn text into vectors with the hosted embedding model, batch them efficiently, and store them in a VectorDB index that matches.
Reranking
Two-stage retrieval with the hosted reranker — fetch a wide candidate set by vector, then reorder it by relevance before answering.