A single Anthropic Messages gateway that runs the server-owned web-search loop through Praxis core’s iterative_request_router (IRR), serves BOTH streaming and buffered clients from one pipeline, and adapts to BOTH backend wire formats from one config
Routes native Anthropic Messages API traffic (/v1/messages and /v1/messages/count_tokens) to a vLLM backend that natively serves the Anthropic Messages API, WITHOUT any request or response body translation
Routes Anthropic Messages API requests to a native /v1/messages backend
Transforms Anthropic Messages API requests and responses for Chat Completions-compatible inference backends
Translates native Anthropic Messages API traffic into OpenAI Chat Completions for a vLLM backend that serves /v1/chat/completions, with the same three-boundary credential isolation as the native passthrough config
Rejects empty, malformed, or non-object JSON request bodies
Routes traffic by classifier-promoted headers so a single listener handles Anthropic Messages, OpenAI Chat Completions, and OpenAI Responses requests
A scoped-credentials variant of full-flow-agentic.yaml