feat: implement decision-based routing with plugin architecture

Signed-off-by: bitliu <[email protected]>

Author: Xunzhuo

src/semantic-router/cmd/main.go

@@ -121,10 +121,8 @@ func main() {
 	logging.Infof("Starting vLLM Semantic Router ExtProc with config: %s", *configPath)
 
 	// Initialize embedding models if configured (Long-context support)
-	cfg, err = config.Load(*configPath)
-	if err != nil {
-		logging.Warnf("Failed to load config for embedding models: %v", err)
-	} else if cfg.Qwen3ModelPath != "" || cfg.GemmaModelPath != "" {
+	// Use the already loaded config instead of calling config.Load() again
+	if cfg.Qwen3ModelPath != "" || cfg.GemmaModelPath != "" {
 		logging.Infof("Initializing embedding models...")
 		logging.Infof("  Qwen3 model: %s", cfg.Qwen3ModelPath)
 		logging.Infof("  Gemma model: %s", cfg.GemmaModelPath)

src/semantic-router/pkg/apiserver/server.go

@@ -17,10 +17,11 @@ import (
 
 // Init starts the API server
 func Init(configPath string, port int, enableSystemPromptAPI bool) error {
-	// Load configuration
-	cfg, err := config.Load(configPath)
-	if err != nil {
-		return fmt.Errorf("failed to load config: %w", err)
+	// Get the global configuration instead of loading from file
+	// This ensures we use the same config as the rest of the application
+	cfg := config.Get()
+	if cfg == nil {
+		return fmt.Errorf("configuration not initialized")
 	}
 
 	// Create classification service - try to get global service with retry
@@ -39,7 +40,7 @@ func Init(configPath string, port int, enableSystemPromptAPI bool) error {
 	}
 
 	// Initialize batch metrics configuration
-	if cfg != nil && cfg.API.BatchClassification.Metrics.Enabled {
+	if cfg.API.BatchClassification.Metrics.Enabled {
 		metricsConfig := metrics.BatchMetricsConfig{
 			Enabled:                   cfg.API.BatchClassification.Metrics.Enabled,
 			DetailedGoroutineTracking: cfg.API.BatchClassification.Metrics.DetailedGoroutineTracking,