// Routes: POST /ai/azure/* (405, 422, dry runs, 503 without a credential) #include #include "support/Assert.hpp" using namespace molo::example::test; int main() { setenv("MOLO_ENV_FILES", "off", 1); for (const char* v : {"AZURE_OPENAI_ENDPOINT", "AZURE_OPENAI_API_KEY", "AZURE_OPENAI_API_VERSION", "AZURE_TENANT_ID", "AZURE_CLIENT_ID", "AZURE_CLIENT_SECRET", "AZURE_AI_ENDPOINT", "AZURE_CONTENT_SAFETY_ENDPOINT", "AZURE_CONTENT_SAFETY_KEY"}) unsetenv(v); Fixture f; auto post = [&](const std::string& route, const std::string& body) { return f.client->post("/ai/azure/" + route, body, "application/json"); }; const char* all[] = {"chat", "embeddings", "images", "speech", "models", "foundry", "safety", "entra", "realtime", "call"}; for (const char* r : all) { std::string route = std::string("/ai/azure/") + r; check(f.client->get(route).status == 405, route + " get is 405"); check(post(r, "not json").status == 422, route + " bad json is 422"); check(post(r, "[1]").status == 422, route + " non-object is 422"); } // 422: input rules run before anything else check(post("chat", "{}").status == 422, "chat needs deployment"); check(post("chat", "{\"deployment\":\"gpt4o\"}").status == 422, "chat needs prompt"); check(post("chat", "{\"deployment\":\"bad name\",\"prompt\":\"hi\"}").status == 422, "bad deployment name"); check(post("chat", "{\"deployment\":\"" + std::string(65, 'a') + "\",\"prompt\":\"hi\"}").status == 422, "deployment over 64"); check(post("chat", "{\"deployment\":\"gpt4o\",\"prompt\":\"hi\",\"dialect\":\"soap\"}").status == 422, "bad dialect"); check(post("chat", "{\"deployment\":\"gpt4o\",\"prompt\":\"hi\",\"max_tokens\":\"x\"}").status == 422, "max_tokens type"); check(post("embeddings", "{\"deployment\":\"e\"}").status == 422, "embeddings need input"); check(post("images", "{\"deployment\":\"i\"}").status == 422, "images need a prompt"); check(post("images", "{\"deployment\":\"i\",\"prompt\":\"x\",\"n\":\"two\"}").status == 422, "image n must be a number"); check(post("speech", "{\"deployment\":\"t\",\"input\":\"x\"}").status == 422, "speech needs a voice"); check(post("foundry", "{}").status == 422, "foundry needs an action"); check(post("foundry", "{\"action\":\"fly\"}").status == 422, "bad foundry action"); check(post("foundry", "{\"action\":\"chat\"}").status == 422, "foundry chat needs a prompt"); check(post("foundry", "{\"action\":\"embeddings\"}").status == 422, "foundry embeddings need input"); check(post("safety", "{\"action\":\"text\"}").status == 422, "safety text needs text"); check(post("safety", "{\"action\":\"shield\"}").status == 422, "shield needs a prompt or documents"); check(post("safety", "{\"action\":\"fly\"}").status == 422, "bad safety action"); check(post("realtime", "{\"action\":\"url\"}").status == 422, "realtime url needs a deployment"); check(post("realtime", "{\"action\":\"url\",\"deployment\":\"bad name\"}").status == 422, "realtime bad deployment"); check(post("realtime", "{\"action\":\"fly\"}").status == 422, "bad realtime action"); check(post("call", "{}").status == 422, "call needs an operation"); check(post("call", "{\"operation\":\"NoSuchCall\"}").status == 422, "unknown operation"); check(post("call", "{\"operation\":\"CreateFile\"}").status == 422, "multipart operation is refused"); check(post("call", "{\"operation\":\"RetrieveFile\",\"ids\":[\"../x\"]}").status == 422, "path traversal id"); check(post("call", "{\"operation\":\"RetrieveFile\"}").status == 422, "missing id"); check(post("call", "{\"operation\":\"CreateConversation\",\"body\":{\"bogus\":1}}").status == 422, "field not in the spec"); check(post("call", "{\"operation\":\"ListContainers\",\"query\":{\"zzz\":\"1\"}}").status == 422, "query not in the spec"); check(post("call", "{\"operation\":\"ListContainers\",\"query\":{\"limit\":5}}").status == 422, "query values are strings"); // dry runs: paid or resource-changing routes show the exact request auto dry = post("chat", "{\"deployment\":\"gpt4o\",\"prompt\":\"hi\"}"); check(dry.status == 200 && contains(dry.body, "\"dry_run\":true") && contains(dry.body, "/openai/v1/chat/completions") && contains(dry.body, "\"model\":\"gpt4o\""), "chat is a dry run on the v1 api"); check(contains(post("chat", "{\"deployment\":\"gpt41\",\"prompt\":\"hi\",\"dialect\":\"responses\"}").body, "/openai/v1/responses"), "responses dialect preview"); check(contains(post("chat", "{\"deployment\":\"Phi-4\",\"prompt\":\"hi\",\"dialect\":\"foundry\"}").body, "/models/chat/completions?api-version=2024-05-01-preview"), "foundry dialect preview"); check(contains(post("embeddings", "{\"deployment\":\"e\",\"input\":\"hi\"}").body, "/openai/v1/embeddings"), "embeddings dry run"); check(contains(post("images", "{\"deployment\":\"i\",\"prompt\":\"a cat\"}").body, "/openai/v1/images/generations"), "images dry run"); check(contains(post("speech", "{\"deployment\":\"t\",\"voice\":\"alloy\",\"input\":\"hi\"}").body, "/openai/v1/audio/speech"), "speech dry run"); check(contains(post("foundry", "{\"action\":\"chat\",\"prompt\":\"hi\"}").body, "\"dry_run\":true"), "foundry chat dry run"); check(contains(post("safety", "{\"action\":\"text\",\"text\":\"hello\"}").body, "/contentsafety/text:analyze?api-version=2024-09-01"), "safety text dry run"); check(contains(post("safety", "{\"action\":\"protected\",\"text\":\"hello\"}").body, "detectProtectedMaterial"), "protected dry run"); check(contains(post("realtime", "{\"action\":\"secret\",\"deployment\":\"rt\",\"seconds\":60}").body, "/openai/v1/realtime/client_secrets"), "client secret dry run"); auto del = post("call", "{\"operation\":\"DeleteFile\",\"ids\":[\"file-1\"]}"); check(del.status == 200 && contains(del.body, "\"dry_run\":true") && contains(del.body, "DELETE /openai/v1/files/file-1"), "delete is a dry run"); check(contains(post("call", "{\"operation\":\"CreateVectorStore\",\"body\":{\"name\":\"docs\"}}").body, "\"dry_run\":true"), "create is a dry run"); check(contains(post("call", "{\"operation\":\"CreateAssistant\",\"body\":{\"model\":\"gpt4o\"}}").body, "/openai/assistants?api-version=2025-04-01-preview"), "dated-only operation shows the fallback api version"); check(contains(post("call", "{\"operation\":\"CreateChatCompletion\",\"deployment\":\"gpt4o\",\"body\":{\"messages\":[]}}").body, "/openai/v1/chat/completions"), "generic chat preview"); // 503: free calls and confirmed calls need the credential for (const char* r : {"models"}) { auto res = post(r, "{}"); check(res.status == 503 && contains(res.body, "not_configured"), std::string(r) + " without key is 503"); } check(post("entra", "{}").status == 503, "entra check without client credentials is 503"); check(post("realtime", "{\"action\":\"url\",\"deployment\":\"rt\"}").status == 503, "realtime url without an endpoint is 503"); check(post("foundry", "{\"action\":\"info\"}").status == 503, "foundry info without key is 503"); check(post("safety", "{\"action\":\"blocklists\"}").status == 503, "blocklists without key is 503"); check(post("call", "{\"operation\":\"ListVectorStores\"}").status == 503, "generic GET without key is 503"); check(post("chat", "{\"deployment\":\"gpt4o\",\"prompt\":\"hi\",\"confirm\":true}").status == 503, "confirmed chat without key is 503"); check(post("embeddings", "{\"deployment\":\"e\",\"input\":\"hi\",\"confirm\":true}").status == 503, "confirmed embeddings without key is 503"); check(post("images", "{\"deployment\":\"i\",\"prompt\":\"x\",\"confirm\":true}").status == 503, "confirmed images without key is 503"); check(post("speech", "{\"deployment\":\"t\",\"voice\":\"alloy\",\"input\":\"x\",\"confirm\":true}").status == 503, "confirmed speech without key is 503"); check(post("safety", "{\"action\":\"text\",\"text\":\"hi\",\"confirm\":true}").status == 503, "confirmed safety without key is 503"); check(post("call", "{\"operation\":\"DeleteFile\",\"ids\":[\"file-1\"],\"confirm\":true}").status == 503, "confirmed delete without key is 503"); check(post("realtime", "{\"action\":\"secret\",\"deployment\":\"rt\",\"confirm\":true}").status == 503, "confirmed client secret without key is 503"); return finish("ai/azure/routes"); }