OpenAI Responses for Codex
Connect Codex through the stateless Responses compatibility endpoint.
POST /v1/responses accepts the stateless Responses wire format used by Codex custom providers. It translates message input and function tools into Strix's existing routing pipeline, so filtering, shared memory, savings evidence, billing, and request logs stay consistent with Chat Completions.
model = "strix-auto"
model_provider = "strix"
[model_providers.strix]
name = "Strix"
base_url = "https://api.strixgate.dev/v1"
env_key = "STRIX_API_KEY"
wire_api = "responses"
http_headers = { "X-Strix-Client" = "codex" }export STRIX_API_KEY="your-strix-key"
codexBoth streaming and non-streaming responses are supported, including text and function-call output items. instructions, message input, function-call results, tool choice, parallel tool calls, output-token limits, reasoning effort, prompt cache keys, and JSON response formats are mapped onto the routed request.
Strix does not store upstream Response objects. Do not send previous_response_id; include the prior response items in input, which is Codex's normal stateless flow. Function tools are supported. Provider-hosted Responses tools and namespace tools are not forwarded through the Chat Completions compatibility layer.
Every successful request exposes the same x-strix-* routing, memory, token, and savings headers documented in routing evidence.