Dynamic Configuration Management for Spring AI Alibaba with Nacos

When building Java AI applications with Spring AI, developers often face challenges such as protecting API‑KEYs, updating Prompt templates on the fly, and managing model parameters or external service credentials securely. The article demonstrates how Spring AI Alibaba combined with Nacos solves these problems.

Connecting Spring AI Alibaba to Nacos

Add a few lines to application.properties to point Spring Cloud to a Nacos server, specify the namespace, and provide access credentials. Example configuration:

spring.cloud.nacos.config.server-addr=${NAOCS_SERVER_ADDR}
spring.cloud.nacos.config.namespace=
spring.cloud.nacos.config.access-key=${NACOS_CONFIG_ACCESS_KEY}
spring.cloud.nacos.config.secret-key=${NACOS_CONFIG_SECRET_KEY}
spring.ai.dashscope.api-key=${AI_DASHSCOPE_API_KEY}

After these settings, any configuration stored in Nacos can be fetched by the application at runtime.

Dynamic Prompt Template Loading

Spring AI provides Prompt Templates for rendering prompts based on user input, but the native approach requires static files. Spring AI Alibaba adds a Nacos‑backed dynamic PromptTemplate. Store a JSON array under data ID spring.ai.alibaba.configurable.prompt in the DEFAULT_GROUP:

[
  {
    "name": "模板名称",
    "template": "预置的Prompt模板内容",
    "model": {"key": "value"}
  }
]

When the configuration is updated in the Nacos console, the new template is applied instantly without restarting the service.

Example Controller Using the Dynamic Template

The following Spring Boot controller demonstrates how to retrieve a PromptTemplate from Nacos and invoke the model:

@RestController
@RequestMapping("/ai")
public class PromptTemplateController {

  private final ChatClient chatClient;
  private final ConfigurablePromptTemplateFactory configurablePromptTemplateFactory;

  @Autowired
  public PromptTemplateController(ChatClient.Builder builder,
      ConfigurablePromptTemplateFactory configurablePromptTemplateFactory) {
    this.chatClient = builder.build();
    this.configurablePromptTemplateFactory = configurablePromptTemplateFactory;
  }

  @GetMapping("/prompt-template")
  public AssistantMessage generate(@RequestParam(value = "author") String author) {
    ConfigurablePromptTemplate template = configurablePromptTemplateFactory.create(
        "test-template", "please list the three most famous books by this {author}.");
    Prompt prompt = StringUtils.hasText(author)
        ? template.create(Map.of("author", author))
        : template.create();
    return chatClient.prompt(prompt).call().chatResponse().getResult().getOutput();
  }
}

Updating the PromptTemplate in Nacos instantly changes the response without redeploying.

Hot‑Updating Model Call Parameters

Spring AI Alibaba also exposes model‑specific options (e.g., temperature, topP, function, tool) via a Nacos configuration with data ID spring.ai.alibaba.dashscope.chat.options. By editing the JSON content in the console, developers can adjust model behavior in real time.

Secure Storage of Sensitive Configurations

For highly sensitive items such as API‑KEYs, Nacos can integrate with a KMS encryption plugin. Encrypted configurations are stored with a cipher-kms-aes-256- prefix. Example import in application.properties:

spring.config.import=optional:nacos:cipher-kms-aes-256-encrypted-config.properties?group=DEFAULT_GROUP&refreshEnabled=true
spring.cloud.nacos.config.kms_region_id=cn-zhangjiakou
spring.cloud.nacos.config.kmsVersion=v1.0
spring.ai.dashscope.api-key=${AI_DASHSCOPE_API_KEY}

When the application starts, Spring decrypts the value and makes it available to the AI client. The refreshEnabled=true flag ensures any subsequent changes are hot‑reloaded.

Conclusion

The combination of Spring AI Alibaba and Nacos provides a seamless, cloud‑native solution for managing dynamic prompts, model parameters, and encrypted secrets, eliminating the need for application restarts and improving development agility for Java AI services.