AI创想

标题: agentcope智能体 [打印本页]

---

作者: 米落枫    时间: 3 小时前
标题: agentcope智能体
作者：CSDN博客

1. # 安装
2. pip install agentscope
4. # 或从源码（获取最新特性）git clone https://github.com/agentscope-ai/agentscope.git
5. cd agentscope && pip install-e.
7. 运行实时语音Agent示例：
8. cd examples/agent/realtime_voice_agent
9. python app.py
10. # 访问 http://localhost:8080 进行语音对话
12. 学习文档
13. https://java.agentscope.io/en/multi-agent/multiagent-debate.html

复制代码

python测试

1. 安装
2. pip install deepagents
3. # 或者用 uv（推荐）
4. uv add deepagents
6. 案例使用
7. testDeepAgent1.py
8. from deepagents import create_deep_agent
9. agent = create_deep_agent()
10. result = agent.invoke({"messages":[{"role":"user", "content":"调研LangGraph并写一份摘要"}]})
12. testDeepAgent2.py
13. # pip install -qU deepagents langchain-openai
14. from deepagents import create_deep_agent
15. def get_weather(city: str) -> str:
16.     """Get weather for a given city."""
17.     return f"It's always sunny in {city}!"
19. agent = create_deep_agent(model="openai:gpt-5.4",
20.     tools=[get_weather],
21.     system_prompt="You are a helpful assistant",
22. )# Run the agent
23. agent.invoke({"messages":[{"role":"user", "content":"what is the weather in sf"}]})

复制代码

java测试

1. 案例一
2. <dependency><groupId>io.agentscope</groupId><artifactId>agentscope</artifactId><version>1.0.11</version></dependency>
4. ReActAgent agent = ReActAgent.builder()
5.     .name("Assistant")
6.     .sysPrompt("You are a helpful AI assistant.")
7.     .model(DashScopeChatModel.builder()
8.         .apiKey(System.getenv("DASHSCOPE_API_KEY"))
9.         .modelName("qwen-max")
10.         .build())
11.     .build();
13. Msg response = agent.call(Msg.builder()
14.         .textContent("Hello!")
15.         .build()).block();
16. System.out.println(response.getTextContent());
18. 案例二
19. <dependencies><dependency><groupId>io.agentscope</groupId><artifactId>agentscope</artifactId><version>1.0.9</version></dependency></dependencies>import io.agentscope.core.model.DashScopeChatModel;
21. public class ModelFactory {
23.     public static DashScopeChatModel createQwenModel(){
24.         String apiKey = System.getenv("DASHSCOPE_API_KEY");if(apiKey == null || apiKey.isBlank()){
25.             throw new IllegalStateException("DASHSCOPE_API_KEY is required");}return DashScopeChatModel.builder()
26.                 .apiKey(apiKey)
27.                 .modelName("qwen-plus")
28.                 //.modelName("gpt-4o")
29.                 //.generateOptions(GenerateOptions.builder().temperature(0.2).build())
30.                 .build();}}
32. 简单输出
33. import io.agentscope.core.ReActAgent;import io.agentscope.core.message.Msg;import io.agentscope.core.model.DashScopeChatModel;
35. public class SimpleAgentDemo {
37.     public static void main(String[] args){
38.         DashScopeChatModel model = ModelFactory.createQwenModel();
40.         ReActAgent agent = ReActAgent.builder()
41.                 .name("Nexus")
42.                 .sysPrompt("You are a helpful enterprise assistant.")
43.                 .model(model)
44.                 .build();
45.         Msg response = agent.call(
46.                 Msg.builder()
47.                         .textContent("你好，请介绍一下 AgentScope 是什么")
48.                         .build()).block();if(response != null){
49.             System.out.println(response.getTextContent());}}}
51. stream流输出
52. import io.agentscope.core.message.Msg;import io.agentscope.core.message.TextBlock;import io.agentscope.core.model.DashScopeChatModel;import io.agentscope.core.model.GenerateOptions;import reactor.core.publisher.Flux;import java.util.List;
54. public class StreamDemo {
56.     public Flux<String> stream(String prompt){
57.         DashScopeChatModel model = ModelFactory.createQwenModel();
59.         GenerateOptions options = GenerateOptions.builder()
60.                 .stream(Boolean.TRUE)
61.                 .build();return model.stream(
62.                         List.of(Msg.builder().textContent(prompt).build()),
63.                         List.of(),
64.                         options)
65.                 .map(chatResponse -> chatResponse.getContent().stream()
66.                         .filter(TextBlock.class::isInstance)
67.                         .map(TextBlock.class::cast)
68.                         .map(TextBlock::getText)
69.                         .reduce("", String::concat))
70.                 .filter(text -> text != null &&!text.isBlank());}}
72. @GetMapping(value ="/test", produces = MediaType.TEXT_EVENT_STREAM_VALUE)
73. public Flux<String> test(@RequestParam String message){return qwenResponseComposer.composeTypewriter(message);}
75. 其他例子
76. 千问例子
77. DashScopeChatModel model = DashScopeChatModel.builder()
78.         .apiKey(System.getenv("DASHSCOPE_API_KEY"))
79.         .modelName("qwen3-max")
80.         .enableThinking(true)  // Automatically enables streaming
81.         .defaultOptions(GenerateOptions.builder()
82.                 .thinkingBudget(5000)  // Token budget for thinking
83.                 .build())
84.         .build();
86. ollama例子
87. OllamaChatModel model =
88.         OllamaChatModel.builder()
89.                 .modelName("qwen3-max")
90.                 .baseUrl("http://localhost:11434")
91.                 .defaultOptions(OllamaOptions.builder()
92.                         .thinkOption(ThinkOption.ThinkBoolean.ENABLED)
93.                         .temperature(0.8)
94.                         .build())
95.                 .build();
97. openai例子
98. OpenAIChatModel model = OpenAIChatModel.builder()
99.         .apiKey(System.getenv("OPENAI_API_KEY"))
100.         .modelName("gpt-4o")
101.         .build();
103. deepseek例子
104. OpenAIChatModel model = OpenAIChatModel.builder()
105.         .apiKey("your-api-key")
106.         .modelName("deepseek-chat")
107.         .baseUrl("https://api.deepseek.com")
108.         .build();
110. claude例子
111. AnthropicChatModel model = AnthropicChatModel.builder()
112.         .apiKey(System.getenv("ANTHROPIC_API_KEY"))
113.         .modelName("claude-sonnet-4-5-20250929")  // Default
114.         .build();
116. Gemini例子
117. GeminiChatModel model = GeminiChatModel.builder()
118.         .apiKey(System.getenv("GEMINI_API_KEY"))
119.         .modelName("gemini-2.5-flash")  // Default
120.         .baseUrl("https://your-gateway.example")  // Optional
121.         .build();
123. Vertex例子
124. GeminiChatModel model = GeminiChatModel.builder()
125.         .modelName("gemini-2.0-flash")
126.         .project("your-gcp-project")
127.         .location("us-central1")
128.         .vertexAI(true)
129.         .credentials(GoogleCredentials.getApplicationDefault())
130.         .build();
132. 多智能体写作// 创建模型
133. DashScopeChatModel model = DashScopeChatModel.builder
134. .apiKey(apiKey)
135. .modelName("qwen-plus")
136. .build;
137. // 创建多个不同角色的智能体
138. ReActAgent planner = ReActAgent.builder
139. .name("规划师")
140. .sysPrompt("你是一个项目规划专家，擅长制定计划和分解任务。")
141. .model(model)
142. .build;
143. ReActAgent executor = ReActAgent.builder
144. .name("执行者")
145. .sysPrompt("你是一个执行专家，擅长将计划转化为具体的执行步骤。")
146. .model(model)
147. .build;
148. ReActAgent reviewer = ReActAgent.builder
149. .name("审查者")
150. .sysPrompt("你是一个审查专家，擅长评估和优化方案。")
151. .model(model)
152. .build;
153. System.out.println("\n[多智能体协作]");
154. String task ="设计一个简单的待办事项管理系统";
155. // 第一步: 规划师制定计划
156. System.out.println("任务: "+ task);
157. System.out.println("\n--- 规划师制定计划 ---");
158. Msg planRequest = Msg.builder
159. .textContent("请为以下任务制定一个详细的计划: "+ task)
160. .build;
161. Msg plan = planner.call(planRequest).block;
162. logger.info("规划师: {}", plan.getTextContent);
163. System.out.println(plan.getTextContent);
164. // 第二步: 执行者细化执行步骤
165. Msg executionRequest = Msg.builder
166. .textContent("基于以下计划，请提供具体的执行步骤:\n"+ plan.getTextContent)
167. .build;
168. Msg execution = executor.call(executionRequest).block;
169. logger.info("执行者: {}", execution.getTextContent);
170. System.out.println(execution.getTextContent);
172. // 第三步: 审查者评估方案
174. Msg reviewRequest = Msg.builder
176. .textContent("请评估以下执行方案，并提出改进建议:\n"+ execution.getTextContent)
178. .build;
180. Msg review = reviewer.call(reviewRequest).block;
182. logger.info("审查者: {}", review.getTextContent);
184. System.out.println(review.getTextContent);
186. 多智能体模型案例
187. // DashScope multi-agent
188. DashScopeChatModel model = DashScopeChatModel.builder()
189.         .apiKey(System.getenv("DASHSCOPE_API_KEY"))
190.         .modelName("qwen3-max")
191.         .formatter(new DashScopeMultiAgentFormatter())
192.         .build();
194. // OpenAI multi-agent
195. OpenAIChatModel model = OpenAIChatModel.builder()
196.         .apiKey(System.getenv("OPENAI_API_KEY"))
197.         .modelName("gpt-4o")
198.         .formatter(new OpenAIMultiAgentFormatter())
199.         .build();
201. // Anthropic multi-agent
202. AnthropicChatModel model = AnthropicChatModel.builder()
203.         .apiKey(System.getenv("ANTHROPIC_API_KEY"))
204.         .formatter(new AnthropicMultiAgentFormatter())
205.         .build();
207. // Gemini multi-agent
208. GeminiChatModel model = GeminiChatModel.builder()
209.         .apiKey(System.getenv("GEMINI_API_KEY"))
210.         .formatter(new GeminiMultiAgentFormatter())
211.         .build();
213. // Ollama multi-agent
214. OllamaChatModel model = OllamaChatModel.builder()
215.         .modelName("qwen3-max")
216.         .formatter(new OllamaMultiAgentFormatter())
217.         .build();
219. 调用工具
220. import io.agentscope.core.ReActAgent;import io.agentscope.core.tool.Toolkit;import io.agentscope.core.model.DashScopeChatModel;
222. // Create model
223. DashScopeChatModel model = DashScopeChatModel.builder()
224.         .apiKey(System.getenv("DASHSCOPE_API_KEY"))
225.         .modelName("qwen-plus")
226.         .build();
228. Toolkit toolkit = new Toolkit();
229. toolkit.registration()
230.         .subAgent(()-> ReActAgent.builder()
231.                 .name("Expert")
232.                 .sysPrompt("You are a domain expert responsible for answering professional questions.")
233.                 .model(model)
234.                 .build())
235.         .apply();
237. // Create main agent with toolkit
238. ReActAgent mainAgent = ReActAgent.builder()
239.         .name("Coordinator")
240.         .sysPrompt("You are a coordinator. When facing professional questions, call the call_expert tool to consult the expert.")
241.         .model(model)
242.         .toolkit(toolkit)
243.         .build();
245. // Main agent will automatically call expert agent when needed
246. Msg response = mainAgent.call(userMsg).block();
248. // 1. Create knowledge base
249. EmbeddingModel embeddingModel = DashScopeTextEmbedding.builder()
250.     .apiKey(System.getenv("DASHSCOPE_API_KEY"))
251.     .modelName("text-embedding-v3")
252.     .dimensions(1024)
253.     .build();
255. Knowledge knowledge = SimpleKnowledge.builder()
256.     .embeddingModel(embeddingModel)
257.     .embeddingStore(InMemoryStore.builder().dimensions(1024).build())
258.     .build();
260. // 2. Add documents
261. TextReader reader = new TextReader(512, SplitStrategy.PARAGRAPH, 50);
262. List<Document> docs = reader.read(ReaderInput.fromString("Text content...")).block();
263. knowledge.addDocuments(docs).block();
265. // 3. Retrieve
266. List<Document> results = knowledge.retrieve("query",
267.     RetrieveConfig.builder().limit(3).scoreThreshold(0.5).build()).block();

复制代码

原文地址：https://blog.csdn.net/qq_35911118/article/details/160597964

---

欢迎光临 AI创想 (http://llms-ai.com/)

Powered by Discuz! X3.4