1-Intro
高层服务. 专注于业务逻辑，而不是低层细节. 类似 Spring Data JPA, Retrofit 这样的框架.
- 采用声明式的接口 来定义 ;
- 使用代理的模式实现接口 ;
- 作为服务层的一个组件 ;
同时支持 LLM 的常见升天, Chat Memory, Tools, RAG 这种。
原理，就是用反射创建一个 动态实现, 属于初级使用.
graph LR A[用户接口定义] --> B[AiServices创建代理] B --> C[代理对象] C --> D[输入转换] D --> E[调用模型] E --> F[输出转换] F --> G[返回结果]
2-Quick start
1)-组装
@Bean
open fun chatLanguageModel(llmConfigProps: LlmConfigProps): ChatLanguageModel {
val supplierType = llmConfigProps.active
return when (supplierType) {
LlmSupplierType.AZURE -> AzureOpenAiChatModel.builder()
.apiKey(llmConfigProps.activeConfig.appKey)
.endpoint(llmConfigProps.activeConfig.endpoint)
.deploymentName(llmConfigProps.activeConfig.modelName)
.logRequestsAndResponses(true)
.build()
else -> throw IllegalArgumentException("Unknown LLM supplier type: $supplierType")
}
}
@Bean
open fun assistant(
chatLanguageModel: ChatLanguageModel,
chatSessionDao: IChatSessionDao,
redisChatMemoryStore: RedisChatMemoryStore
): Assistant {
return AiServices.builder(Assistant::class.java)
.systemMessageProvider {
chatSessionDao.findById(it.toString().toInt()).map(ChatSession::systemMessage).orElse(null)
}.chatMemoryProvider {
MessageWindowChatMemory.builder()
.id(it)
.maxMessages(30)
.chatMemoryStore(redisChatMemoryStore)
.build()
}
.chatLanguageModel(chatLanguageModel)
.build()
}2)-定义基本的 chat 函数
interface Assistant {
fun chat(@MemoryId memoryId: Int, @UserMessage userMessage: String): String
}3)-实现 redis chat memory
@Component
open class RedisChatMemoryStore(
private val redisTemplate: RedisTemplate<String, String>,
@Value("\${chat.memory.prefix}")
private val prefix: String
) : ChatMemoryStore, ILogging {
fun key(memoryId: Any): String {
return "$prefix$memoryId"
}
override fun getMessages(memoryId: Any): List<ChatMessage> {
val key = key(memoryId)
logger.info("Retrieving messages from Redis, key:{}", key)
val json = redisTemplate.opsForValue().get(key)
return ChatMessageDeserializer.messagesFromJson(json)
}
override fun updateMessages(
memoryId: Any,
messages: List<ChatMessage>
) {
val key = key(memoryId)
logger.info("Updating messages in Redis, key: {}, size:{}", key, messages.size)
val json = ChatMessageSerializer.messagesToJson(messages)
/*默认保留30天*/
redisTemplate.opsForValue().set(key, json, 30L, TimeUnit.DAYS)
}
override fun deleteMessages(memoryId: Any) {
val key = key(memoryId)
logger.info("Deleting messages from Redis, key : {}", key)
redisTemplate.opsForValue().getOperations().delete(key)
}
}