Read Tool & Reasoning Tokens from GenAI API instead of Catalog Metadata

sdk_v2/cpp/src/inferencing/generative/genai_model_instance.cc

@@ -111,6 +111,15 @@ OgaModel& GenAIModelInstance::GetOgaModel() {
   return *oga_model_;
 }
 
+std::string GenAIModelInstance::GetTag(const char* tag_name) const {
+  if (!oga_model_) {
+    return {};
+  }
+  OgaString tag = oga_model_->GetTag(tag_name);
+  const char* p = tag;
+  return p ? std::string(p) : std::string();
+}
+
 OgaTokenizer& GenAIModelInstance::GetOgaTokenizer() {
   if (!tokenizer_) {
     FL_THROW(FOUNDRY_LOCAL_ERROR_INTERNAL, "OGA tokenizer is null");

sdk_v2/cpp/src/inferencing/generative/genai_model_instance.h

@@ -36,6 +36,11 @@ class GenAIModelInstance {
   ExecutionProvider EP() const { return ep_; }
   bool IsMultiModal() const;
 
+  /// Query a tag from the GenAI model (reads genai_config.json with model-family fallback).
+  /// Returns the tag value or empty string if not found.  Supported keys:
+  /// "tool_call_start", "tool_call_end", "reasoning_start", "reasoning_end".
+  std::string GetTag(const char* tag_name) const;
+
   /// Access the underlying OGA objects (for future chat generation work).
   OgaModel& GetOgaModel();
   OgaTokenizer& GetOgaTokenizer();

sdk_v2/cpp/src/model.cc

@@ -218,6 +218,37 @@ void Model::Load(ExecutionProvider ep) {
   if (result.status == ModelLoadManager::LoadStatus::kModelNotFound) {
     FL_THROW(FOUNDRY_LOCAL_ERROR_INTERNAL, "model not found at path: " + local_path_);
   }
+
+  // Enrich ModelInfo with metadata from the loaded GenAI model (genai_config.json + fallback map).
+  // This makes tool/reasoning tags available via ModelInfo regardless of whether they came from
+  // catalog metadata, so downstream code doesn't need to know about multiple metadata sources.
+  if (result.model && result.status == ModelLoadManager::LoadStatus::kSuccess) {
+    auto enrich = [&](const char* tag_name, const char* prop_key) {
+      std::string val = result.model->GetTag(tag_name);
+      if (!val.empty() && !info_.GetPropertyStr(prop_key)) {
+        info_.string_properties[prop_key] = std::move(val);
+      }
+    };
+
+    enrich("tool_call_start", FOUNDRY_LOCAL_MODEL_PROP_TOOL_CALL_START_STR);
+    enrich("tool_call_end", FOUNDRY_LOCAL_MODEL_PROP_TOOL_CALL_END_STR);
+    enrich("reasoning_start", FOUNDRY_LOCAL_MODEL_PROP_REASONING_START_STR);
+    enrich("reasoning_end", FOUNDRY_LOCAL_MODEL_PROP_REASONING_END_STR);
+
+    // Infer support flags from the presence of start tokens
+    if (!info_.GetPropertyInt(FOUNDRY_LOCAL_MODEL_PROP_SUPPORTS_TOOL_CALLING_INT)) {
+      const auto* tc = info_.GetPropertyStr(FOUNDRY_LOCAL_MODEL_PROP_TOOL_CALL_START_STR);
+      if (tc && !tc->empty()) {
+        info_.int_properties[FOUNDRY_LOCAL_MODEL_PROP_SUPPORTS_TOOL_CALLING_INT] = 1;
+      }
+    }
+    if (!info_.GetPropertyInt(FOUNDRY_LOCAL_MODEL_PROP_SUPPORTS_REASONING_INT)) {
+      const auto* rs = info_.GetPropertyStr(FOUNDRY_LOCAL_MODEL_PROP_REASONING_START_STR);
+      if (rs && !rs->empty()) {
+        info_.int_properties[FOUNDRY_LOCAL_MODEL_PROP_SUPPORTS_REASONING_INT] = 1;
+      }
+    }
+  }
 }
 
 void Model::Unload() {