cd backend && go test ./...npm --prefix frontend testStart the router backend, then point local Codex CLI to the router:
model_provider = "router"
[model_providers.router]
name = "router"
base_url = "http://127.0.0.1:6789/ai-router/api"
wire_api = "responses"
requires_openai_auth = trueMinimal checks:
curl http://127.0.0.1:6789/ai-router/api/models- Send one non-stream request to
POST /ai-router/api/responses - Send one stream request to
POST /ai-router/api/responsesand verify the stream terminates with an upstream-aligned terminal event - Switch between two official
auth.jsonaccounts and verify requests do not hang or lose terminal output - From Codex CLI, run one short prompt and verify the router account list shows a run against the active account
Thin gateway notes:
- Third-party smoke tests are valid only for providers that natively implement
/responses. - Do not expect gateway-synthesized response retrieval endpoints to be available.
- Treat upstream
response_idas authoritative.