feat: support configurable execution mode (CLI or API)

internal/adapter/checkstyle/converter.go

@@ -121,17 +121,17 @@ func (c *Converter) ConvertRules(ctx context.Context, rules []schema.UserRule, l
 	// Separate modules into Checker-level and TreeWalker-level
 	// Checker-level modules (NOT under TreeWalker)
 	checkerLevelModules := map[string]bool{
-		"LineLength":                        true,
-		"FileLength":                        true,
-		"FileTabCharacter":                  true,
-		"NewlineAtEndOfFile":                true,
-		"UniqueProperties":                  true,
-		"OrderedProperties":                 true,
-		"Translation":                       true,
-		"SuppressWarningsFilter":            true,
+		"LineLength":                         true,
+		"FileLength":                         true,
+		"FileTabCharacter":                   true,
+		"NewlineAtEndOfFile":                 true,
+		"UniqueProperties":                   true,
+		"OrderedProperties":                  true,
+		"Translation":                        true,
+		"SuppressWarningsFilter":             true,
 		"BeforeExecutionExclusionFileFilter": true,
-		"SuppressionFilter":                 true,
-		"SuppressionCommentFilter":          true,
+		"SuppressionFilter":                  true,
+		"SuppressionCommentFilter":           true,
 	}
 
 	var checkerModules []checkstyleModule
@@ -255,8 +255,8 @@ Output:
 
 	userPrompt := fmt.Sprintf("Convert this Java rule to Checkstyle module:\n\n%s", rule.Say)
 
-	// Call LLM with power model + low reasoning
-	response, err := llmClient.Request(systemPrompt, userPrompt).WithPower(llm.ReasoningMinimal).Execute(ctx)
+	// Call LLM with minimal complexity
+	response, err := llmClient.Request(systemPrompt, userPrompt).WithComplexity(llm.ComplexityMinimal).Execute(ctx)
 	if err != nil {
 		return nil, fmt.Errorf("LLM call failed: %w", err)
 	}

internal/adapter/eslint/converter.go

@@ -217,8 +217,8 @@ Output:
 		userPrompt += fmt.Sprintf("\nSeverity: %s", rule.Severity)
 	}
 
-	// Call LLM with power model + low reasoning
-	response, err := llmClient.Request(systemPrompt, userPrompt).WithPower(llm.ReasoningMinimal).Execute(ctx)
+	// Call LLM with minimal complexity
+	response, err := llmClient.Request(systemPrompt, userPrompt).WithComplexity(llm.ComplexityMinimal).Execute(ctx)
 	if err != nil {
 		return "", nil, fmt.Errorf("LLM call failed: %w", err)
 	}

internal/adapter/pmd/converter.go

@@ -188,8 +188,8 @@ IMPORTANT: Return ONLY the JSON object. Do NOT include description, message, or
 
 	userPrompt := fmt.Sprintf("Convert this Java rule to PMD rule reference:\n\n%s", rule.Say)
 
-	// Call LLM with power model + low reasoning
-	response, err := llmClient.Request(systemPrompt, userPrompt).WithPower(llm.ReasoningMinimal).Execute(ctx)
+	// Call LLM with minimal complexity
+	response, err := llmClient.Request(systemPrompt, userPrompt).WithComplexity(llm.ComplexityMinimal).Execute(ctx)
 	if err != nil {
 		return nil, fmt.Errorf("LLM call failed: %w", err)
 	}

internal/adapter/prettier/converter.go

@@ -134,7 +134,7 @@ Output:
 	userPrompt := fmt.Sprintf("Convert this rule to Prettier configuration:\n\n%s", rule.Say)
 
 	// Call LLM
-	response, err := llmClient.Request(systemPrompt, userPrompt).WithPower(llm.ReasoningMinimal).Execute(ctx)
+	response, err := llmClient.Request(systemPrompt, userPrompt).WithComplexity(llm.ComplexityMinimal).Execute(ctx)
 	if err != nil {
 		return nil, fmt.Errorf("LLM call failed: %w", err)
 	}

internal/adapter/pylint/converter.go

@@ -210,7 +210,7 @@ Output:
 	}
 
 	// Call LLM
-	response, err := llmClient.Request(systemPrompt, userPrompt).WithPower(llm.ReasoningMinimal).Execute(ctx)
+	response, err := llmClient.Request(systemPrompt, userPrompt).WithComplexity(llm.ComplexityMinimal).Execute(ctx)
 	if err != nil {
 		return "", nil, fmt.Errorf("LLM call failed: %w", err)
 	}

internal/adapter/tsc/converter.go

@@ -148,7 +148,7 @@ Output:
 	userPrompt := fmt.Sprintf("Convert this rule to TypeScript compiler configuration:\n\n%s", rule.Say)
 
 	// Call LLM
-	response, err := llmClient.Request(systemPrompt, userPrompt).WithPower(llm.ReasoningMinimal).Execute(ctx)
+	response, err := llmClient.Request(systemPrompt, userPrompt).WithComplexity(llm.ComplexityMinimal).Execute(ctx)
 	if err != nil {
 		return nil, fmt.Errorf("LLM call failed: %w", err)
 	}

internal/cmd/api_key.go

@@ -16,11 +16,6 @@ func promptAPIKeySetup() {
 	promptAPIKeyConfiguration(false)
 }
 
-// promptAPIKeyIfNeeded checks if OpenAI API key is configured and prompts if not
-func promptAPIKeyIfNeeded() {
-	promptAPIKeyConfiguration(true)
-}
-
 // promptAPIKeyConfiguration handles API key configuration with optional existence check
 func promptAPIKeyConfiguration(checkExisting bool) {
 	envPath := filepath.Join(".sym", ".env")

internal/cmd/convert.go

@@ -40,6 +40,8 @@ map them to appropriate linter rules.`,
   # Convert for specific linter
   sym convert -i user-policy.json --targets eslint
 
+  # Convert for Java with specific model
+  sym convert -i user-policy.json --targets checkstyle,pmd --openai-model gpt-5-mini
   # Convert for Java with specific model
   sym convert -i user-policy.json --targets checkstyle,pmd --openai-model gpt-5-mini
 
@@ -131,18 +133,19 @@ func runNewConverter(userPolicy *schema.UserPolicy) error {
 		convertOutputDir = ".sym"
 	}
 
-	// Setup OpenAI client
-	apiKey, err := getAPIKey()
-	if err != nil {
-		return fmt.Errorf("OpenAI API key required: %w", err)
-	}
-
 	timeout := time.Duration(convertTimeout) * time.Second
 	llmClient := llm.NewClient(
-		apiKey,
 		llm.WithTimeout(timeout),
 	)
 
+	// Ensure at least one backend is available (MCP/CLI/API)
+	availabilityCtx, cancelAvailability := context.WithTimeout(context.Background(), timeout)
+	defer cancelAvailability()
+
+	if err := llmClient.CheckAvailability(availabilityCtx); err != nil {
+		return fmt.Errorf("no available LLM backend for convert: %w\nTip: run 'sym init --setup-llm' or configure LLM_BACKEND / LLM_CLI / OPENAI_API_KEY in .sym/.env", err)
+	}
+
 	// Create new converter
 	conv := converter.NewConverter(llmClient, convertOutputDir)
 

internal/cmd/init.go

@@ -33,19 +33,23 @@ This command:
 }
 
 var (
-	initForce        bool
-	skipMCPRegister  bool
-	registerMCPOnly  bool
-	skipAPIKey       bool
-	setupAPIKeyOnly  bool
+	initForce       bool
+	skipMCPRegister bool
+	registerMCPOnly bool
+	skipAPIKey      bool
+	setupAPIKeyOnly bool
+	skipLLMSetup    bool
+	setupLLMOnly    bool
 )
 
 func init() {
 	initCmd.Flags().BoolVarP(&initForce, "force", "f", false, "Overwrite existing roles.json")
 	initCmd.Flags().BoolVar(&skipMCPRegister, "skip-mcp", false, "Skip MCP server registration prompt")
 	initCmd.Flags().BoolVar(&registerMCPOnly, "register-mcp", false, "Register MCP server only (skip roles/policy init)")
-	initCmd.Flags().BoolVar(&skipAPIKey, "skip-api-key", false, "Skip OpenAI API key configuration prompt")
-	initCmd.Flags().BoolVar(&setupAPIKeyOnly, "setup-api-key", false, "Setup OpenAI API key only (skip roles/policy init)")
+	initCmd.Flags().BoolVar(&skipAPIKey, "skip-api-key", false, "Skip OpenAI API key configuration prompt (deprecated, use --skip-llm)")
+	initCmd.Flags().BoolVar(&setupAPIKeyOnly, "setup-api-key", false, "Setup OpenAI API key only (deprecated, use --setup-llm)")
+	initCmd.Flags().BoolVar(&skipLLMSetup, "skip-llm", false, "Skip LLM backend configuration prompt")
+	initCmd.Flags().BoolVar(&setupLLMOnly, "setup-llm", false, "Setup LLM backend only (skip roles/policy init)")
 }
 
 func runInit(cmd *cobra.Command, args []string) {
@@ -56,13 +60,20 @@ func runInit(cmd *cobra.Command, args []string) {
 		return
 	}
 
-	// API key setup only mode
+	// API key setup only mode (deprecated)
 	if setupAPIKeyOnly {
 		fmt.Println("🔑 Setting up OpenAI API key...")
 		promptAPIKeySetup()
 		return
 	}
 
+	// LLM setup only mode
+	if setupLLMOnly {
+		fmt.Println("🤖 Setting up LLM backend...")
+		promptLLMBackendSetup()
+		return
+	}
+
 	// Check if logged in
 	if !config.IsLoggedIn() {
 		fmt.Println("❌ Not logged in")
@@ -156,9 +167,9 @@ func runInit(cmd *cobra.Command, args []string) {
 		promptMCPRegistration()
 	}
 
-	// API key configuration prompt
-	if !skipAPIKey {
-		promptAPIKeyIfNeeded()
+	// LLM backend configuration prompt
+	if !skipLLMSetup && !skipAPIKey {
+		promptLLMBackendSetup()
 	}
 
 	// Show dashboard guide after all initialization is complete