Transforms Anthropic Messages API requests to Chat Completions-compatible request bodies and transforms compatible responses back. The name refers to the Chat Completions wire shape, not the OpenAI Responses API; any Chat Completions-compatible backend is a valid target, not only OpenAI.
Lowers rich client-owned tool declarations to private function tools for a function-only Responses backend and restores the typed items on the way back.
Resolves file_id and file_url references in Responses API input by fetching content from a Files API or remote URL via ApiClient and inlining the base64-encoded content in the provider-native field.
Validates previous_response_id by fetching the stored response, confirming its status is "completed", and populating ResponsesState with the full conversation history (stored turns + current input).
Parses tool definitions and tool_choice from Responses API request bodies and promotes routing facts to metadata and filter results without mutating the body.
Extracts A2A protocol metadata from JSON-RPC request bodies and promotes method, family, task ID, streaming detection, and version to request headers, filter results, and durable metadata for routing.
Extracts MCP protocol metadata from JSON-RPC request bodies and promotes method, tool/resource/prompt name, JSON-RPC kind, protocol version, and session presence to request headers/filter results; stores session ID in durable metadata.
Calls an external AI guardrail provider to evaluate request and response bodies. The provider determines whether content should be passed, blocked, or redacted.
Token-denominated rate limiter: reserves an estimated cost at admission, reconciles against actual usage after the response completes. Evaluates an ordered list of rules, each with its own optional match condition, algorithm choice, and budget.
Rejects provider-bound requests whose estimated serialized request body or requested output exceeds configured limits. Missing output-limit fields are rejected when max_output_tokens is configured; this strict behavior prevents providers from applying an unbounded model default. At least one of max_input_tokens or max_output_tokens is required, and max_body_bytes defaults to 2 MiB.
Injects Praxis-Token-Input, Praxis-Token-Output, and Praxis-Token-Total headers into downstream responses when token usage data is present in [filter_metadata]. Also injects Praxis-Token-Status when usage capture failed (e.g. overflow), so an unavailable count is never silently indistinguishable from a genuine zero.