Merge pull request #42 from lucaromagnoli/security/remove-mac-key

lucaromagnoli · web-flow · commit f324baa53244 · 2025-08-08T21:31:04.000+01:00
Security/remove mac key
diff --git a/.gitignore b/.gitignore
@@ -53,4 +53,7 @@ docs/build/
 !tests/*_sample.*
 sample_*
 !tests/sample_*
-response_samples/
+response_samples/
+# Never commit signing keys
+llmcpp-mac-key.pem
+
diff --git a/README.md b/README.md
@@ -132,6 +132,7 @@ auto response2 = client.sendRequest(OpenAI::Model::GPT_4_1_Mini, "Summarize this
 ```
 
 **Available models:**
+- GPT_5, GPT_5_Mini, GPT_5_Nano
 - O3, O3_Mini
 - O1, O1_Mini, O1_Preview, O1_Pro
 - O4_Mini, O4_Mini_Deep_Research
@@ -307,14 +308,25 @@ The library provides type-safe model selection using the `OpenAI::Model` enum:
 
 ```cpp
 // Available model enums
-OpenAI::Model::GPT_4_1        // gpt-4.1 - Latest model with superior coding and structured outputs
-OpenAI::Model::GPT_4_1_Mini   // gpt-4.1-mini - Balanced performance and cost
-OpenAI::Model::GPT_4_1_Nano   // gpt-4.1-nano - Fastest and cheapest option
-OpenAI::Model::GPT_4o         // gpt-4o - Good balance of performance and cost
-OpenAI::Model::GPT_4o_Mini    // gpt-4o-mini - Cost-effective for basic tasks
-OpenAI::Model::GPT_4_5        // gpt-4.5 - Preview model (deprecated July 2025)
-OpenAI::Model::GPT_3_5_Turbo  // gpt-3.5-turbo - Legacy model
-OpenAI::Model::Custom         // For custom model names
+OpenAI::Model::GPT_5                    // gpt-5 - Next-generation model
+OpenAI::Model::GPT_5_Mini               // gpt-5-mini - Smaller, cost-effective variant
+OpenAI::Model::GPT_5_Nano               // gpt-5-nano - Fastest and cheapest GPT-5 variant
+OpenAI::Model::O3                       // o3 - Latest reasoning model
+OpenAI::Model::O3_Mini                  // o3-mini - Cost-effective reasoning model
+OpenAI::Model::O1                       // o1 - Advanced reasoning model
+OpenAI::Model::O1_Mini                  // o1-mini - Cost-effective O1 model
+OpenAI::Model::O1_Preview               // o1-preview - Preview version
+OpenAI::Model::O1_Pro                   // o1-pro - Professional version
+OpenAI::Model::O4_Mini                  // o4-mini - Latest mini model
+OpenAI::Model::O4_Mini_Deep_Research    // o4-mini-deep-research - Research focused
+OpenAI::Model::GPT_4_1                  // gpt-4.1 - Latest model with superior coding and structured outputs
+OpenAI::Model::GPT_4_1_Mini             // gpt-4.1-mini - Balanced performance and cost
+OpenAI::Model::GPT_4_1_Nano             // gpt-4.1-nano - Fastest and cheapest option
+OpenAI::Model::GPT_4o                   // gpt-4o - Good balance of performance and cost
+OpenAI::Model::GPT_4o_Mini              // gpt-4o-mini - Cost-effective for basic tasks
+OpenAI::Model::GPT_4_5                  // gpt-4.5 - Preview model (deprecated July 2025)
+OpenAI::Model::GPT_3_5_Turbo            // gpt-3.5-turbo - Legacy model
+OpenAI::Model::Custom                   // For custom model names
 ```
 
 #### Model Selection Helpers
diff --git a/include/openai/OpenAITypes.h b/include/openai/OpenAITypes.h
@@ -18,7 +18,9 @@ namespace OpenAI {
  */
 enum class Model {
     // GPT-5 series (Latest - 2025)
-    GPT_5,  // gpt-5 - Next-generation model
+    GPT_5,       // gpt-5 - Next-generation model
+    GPT_5_Mini,  // gpt-5-mini - Smaller, cost-effective variant
+    GPT_5_Nano,  // gpt-5-nano - Fastest and cheapest GPT-5 variant
 
     // O3 series (Latest - 2025)
     O3,       // o3 - Latest reasoning model
@@ -61,6 +63,10 @@ inline std::string toString(Model model) {
     switch (model) {
         case Model::GPT_5:
             return "gpt-5";
+        case Model::GPT_5_Mini:
+            return "gpt-5-mini";
+        case Model::GPT_5_Nano:
+            return "gpt-5-nano";
         case Model::O3:
             return "o3";
         case Model::O3_Mini:
@@ -102,6 +108,8 @@ inline std::string toString(Model model) {
  */
 inline Model modelFromString(const std::string& modelStr) {
     if (modelStr == "gpt-5") return Model::GPT_5;
+    if (modelStr == "gpt-5-mini") return Model::GPT_5_Mini;
+    if (modelStr == "gpt-5-nano") return Model::GPT_5_Nano;
     if (modelStr == "o3") return Model::O3;
     if (modelStr == "o3-mini") return Model::O3_Mini;
     if (modelStr == "o1") return Model::O1;
@@ -126,6 +134,8 @@ inline Model modelFromString(const std::string& modelStr) {
 inline bool supportsStructuredOutputs(Model model) {
     switch (model) {
         case Model::GPT_5:
+        case Model::GPT_5_Mini:
+        case Model::GPT_5_Nano:
         case Model::O3:
         case Model::O3_Mini:
         case Model::O1:
@@ -1014,8 +1024,9 @@ std::string getRecommendedApiForModel(const std::string& model);
 
 // Model lists for different APIs
 const std::vector<std::string> RESPONSES_MODELS = {
-    "gpt-5",       "gpt-4o", "gpt-4o-mini", "gpt-4.1", "gpt-4.1-nano", "gpt-4.1-mini",
-    "gpt-image-1", "o1",     "o3-mini",     "o3",      "o4-mini",      "computer-use-preview"};
+    "gpt-5",        "gpt-5-mini",          "gpt-5-nano",  "gpt-4o", "gpt-4o-mini", "gpt-4.1",
+    "gpt-4.1-nano", "gpt-4.1-mini",        "gpt-image-1", "o1",     "o3-mini",     "o3",
+    "o4-mini",      "computer-use-preview"};
 
 const std::vector<std::string> CHAT_COMPLETION_MODELS = {"gpt-4", "gpt-4-turbo", "gpt-4o",
                                                          "gpt-4o-mini", "gpt-3.5-turbo"};
diff --git a/llmcpp-mac-key.pem b/llmcpp-mac-key.pem
diff --git a/src/openai/OpenAIClient.cpp b/src/openai/OpenAIClient.cpp
@@ -175,7 +175,7 @@ OpenAI::ResponsesResponse OpenAIClient::sendResponsesRequest(
         try {
             // Reasonable defaults: wait up to 90s, polling every 2s
             response = responsesApi_->waitForCompletion(response.id, /*timeoutSeconds=*/90,
-                                                       /*pollIntervalSeconds=*/2);
+                                                        /*pollIntervalSeconds=*/2);
         } catch (const std::exception& /*e*/) {
             // Fall through and return the last known response (likely non-completed)
         }
@@ -352,7 +352,8 @@ OpenAI::Model OpenAIClient::stringToModel(const std::string& modelStr) {
 }
 
 std::vector<OpenAI::Model> OpenAIClient::getAvailableModelEnums() {
-    return {OpenAI::Model::GPT_5,        OpenAI::Model::GPT_4_1,       OpenAI::Model::GPT_4_1_Mini,
-            OpenAI::Model::GPT_4_1_Nano, OpenAI::Model::GPT_4o,        OpenAI::Model::GPT_4o_Mini,
-            OpenAI::Model::GPT_4_5,      OpenAI::Model::GPT_3_5_Turbo, OpenAI::Model::Custom};
+    return {OpenAI::Model::GPT_5,         OpenAI::Model::GPT_5_Mini,   OpenAI::Model::GPT_5_Nano,
+            OpenAI::Model::GPT_4_1,       OpenAI::Model::GPT_4_1_Mini, OpenAI::Model::GPT_4_1_Nano,
+            OpenAI::Model::GPT_4o,        OpenAI::Model::GPT_4o_Mini,  OpenAI::Model::GPT_4_5,
+            OpenAI::Model::GPT_3_5_Turbo, OpenAI::Model::Custom};
 }
diff --git a/tests/CMakeLists.txt b/tests/CMakeLists.txt
@@ -28,6 +28,7 @@ set(UNIT_TEST_SOURCES
 set(INTEGRATION_TEST_SOURCES
     integration/test_simple_integration.cpp
     integration/test_openai_integration.cpp  # Re-enabling to check errors
+    integration/test_benchmarks.cpp
 )
 
 # Create test executable with all tests
diff --git a/tests/integration/test_benchmarks.cpp b/tests/integration/test_benchmarks.cpp
@@ -0,0 +1,69 @@
+#include <catch2/catch_test_macros.hpp>
+#include <chrono>
+#include <cstdlib>
+#include <iostream>
+#include <string>
+
+#include "openai/OpenAIClient.h"
+#include "openai/OpenAITypes.h"
+
+using namespace std::chrono;
+
+static bool isReasoningModel(OpenAI::Model model) {
+    return model == OpenAI::Model::GPT_5 || model == OpenAI::Model::GPT_5_Mini ||
+           model == OpenAI::Model::GPT_5_Nano || model == OpenAI::Model::O3 ||
+           model == OpenAI::Model::O3_Mini || model == OpenAI::Model::O1 ||
+           model == OpenAI::Model::O1_Mini || model == OpenAI::Model::O1_Preview ||
+           model == OpenAI::Model::O1_Pro || model == OpenAI::Model::O4_Mini;
+}
+
+TEST_CASE("OpenAI model benchmarks (structured outputs)", "[openai][integration][benchmark]") {
+    const char* runBenchEnv = std::getenv("LLMCPP_RUN_BENCHMARKS");
+    if (!runBenchEnv || std::string(runBenchEnv) != "1") {
+        SUCCEED("Benchmarks skipped. Set LLMCPP_RUN_BENCHMARKS=1 to enable.");
+        return;
+    }
+
+    const char* apiKey = std::getenv("OPENAI_API_KEY");
+    REQUIRE(apiKey != nullptr);
+
+    OpenAIClient client(apiKey);
+
+    // Minimal structured output schema
+    json schema = {{"type", "object"},
+                   {"properties", {{"answer", {{"type", "string"}}}}},
+                   {"required", json::array({"answer"})}};
+
+    // Simple input
+    auto input = OpenAI::ResponsesInput::fromText("Reply with the word OK.");
+
+    // Iterate through response-capable models
+    for (const auto& modelName : OpenAI::RESPONSES_MODELS) {
+        DYNAMIC_SECTION("Benchmark model: " << modelName) {
+            OpenAI::ResponsesRequest req;
+            req.model = modelName;
+            req.input = input;
+            req.text = OpenAI::TextOutputConfig("bench_schema", schema, true);
+            req.maxOutputTokens = 16;
+
+            // Tweak reasoning parameters when appropriate
+            auto modelEnum = OpenAI::modelFromString(modelName);
+            if (isReasoningModel(modelEnum)) {
+                req.reasoningEffort = std::string("low");
+            }
+
+            const auto start = steady_clock::now();
+            auto response = client.sendResponsesRequest(req);
+            const auto end = steady_clock::now();
+
+            const auto elapsedMs = duration_cast<milliseconds>(end - start).count();
+            std::cout << "[BENCH] model=" << modelName << ", ms=" << elapsedMs
+                      << ", success=" << (response.isCompleted() && !response.hasError())
+                      << std::endl;
+
+            // Sanity: we should at least get a response object back; don't assert success to avoid
+            // flakes
+            REQUIRE(!response.id.empty());
+        }
+    }
+}

Original file line number	Diff line number	Diff line change
`@@ -175,7 +175,7 @@ OpenAI::ResponsesResponse OpenAIClient::sendResponsesRequest(`
`175`	`175`	`try {`
`176`	`176`	`// Reasonable defaults: wait up to 90s, polling every 2s`
`177`	`177`	`response = responsesApi_->waitForCompletion(response.id, /timeoutSeconds=/90,`
`178`		`- /pollIntervalSeconds=/2);`
	`178`	`+ /pollIntervalSeconds=/2);`
`179`	`179`	`} catch (const std::exception& /e/) {`
`180`	`180`	`// Fall through and return the last known response (likely non-completed)`
`181`	`181`	`}`
`@@ -352,7 +352,8 @@ OpenAI::Model OpenAIClient::stringToModel(const std::string& modelStr) {`
`352`	`352`	`}`
`353`	`353`
`354`	`354`	`std::vector<OpenAI::Model> OpenAIClient::getAvailableModelEnums() {`
`355`		`- return {OpenAI::Model::GPT_5, OpenAI::Model::GPT_4_1, OpenAI::Model::GPT_4_1_Mini,`
`356`		`- OpenAI::Model::GPT_4_1_Nano, OpenAI::Model::GPT_4o, OpenAI::Model::GPT_4o_Mini,`
`357`		`- OpenAI::Model::GPT_4_5, OpenAI::Model::GPT_3_5_Turbo, OpenAI::Model::Custom};`
	`355`	`+ return {OpenAI::Model::GPT_5, OpenAI::Model::GPT_5_Mini, OpenAI::Model::GPT_5_Nano,`
	`356`	`+ OpenAI::Model::GPT_4_1, OpenAI::Model::GPT_4_1_Mini, OpenAI::Model::GPT_4_1_Nano,`
	`357`	`+ OpenAI::Model::GPT_4o, OpenAI::Model::GPT_4o_Mini, OpenAI::Model::GPT_4_5,`
	`358`	`+ OpenAI::Model::GPT_3_5_Turbo, OpenAI::Model::Custom};`
`358`	`359`	`}`
Original file line number	Diff line number	Diff line change
`@@ -28,6 +28,7 @@ set(UNIT_TEST_SOURCES`
`28`	`28`	`set(INTEGRATION_TEST_SOURCES`
`29`	`29`	`integration/test_simple_integration.cpp`
`30`	`30`	`integration/test_openai_integration.cpp # Re-enabling to check errors`
	`31`	`+ integration/test_benchmarks.cpp`
`31`	`32`	`)`
`32`	`33`
`33`	`34`	`# Create test executable with all tests`