重构图像生成和搜索工具；更新主代理来处理输入图像

- 更新了“generate_image.py”以接受输入图像以增强图像生成。
- 修改了`pexels_search.py​​`和`unsplash_search.py​​`以将日志记录和上传路径从“explorer”更改为“explore”。
- 调整了“print_graph”和“sketch_graph”以提取最新的用户输入并处理输入图像以生成打印和草图图像。
- 重构“generate_print_tool”和“generate_sketch_tool”以接受输入图像。
- 更新了“main_agent.py”以包含状态中的输入图像并调整了图形构建过程。
- 增强了“service.py”来管理输入图像并改进了流媒体期间的事件处理。
- 更新了新软件包和版本的“pyproject.toml”和“uv.lock”中的依赖项。

app/service/fashion_agent/graph_node/explore_graph/graph.py

@@ -9,8 +9,8 @@ from langgraph.graph import END, START, StateGraph
 from langgraph.graph.message import add_messages
 from pydantic import BaseModel, Field
 from langchain_core.runnables import RunnableConfig
-from app.service.fashion_agent.graph_node.explorer_graph.tools import explor_tool
-from app.service.fashion_agent.init_llm import build_llm
+from app.service.fashion_agent.graph_node.explore_graph.tools import explore_tool
+from app.service.fashion_agent.init_llm import qwen_plus_llm
 
 logger = logging.getLogger()
 
@@ -18,10 +18,10 @@ logger = logging.getLogger()
 """定义状态"""
 
 
-class ExplorerState(TypedDict):
+class exploreState(TypedDict):
     messages: Required[Annotated[list[AnyMessage], add_messages]]
-    input_text: str
-    search_query: str
+    input_text: str = ""
+    search_query: str = ""
     image_results: list[dict]  # 每项包含 image_url 和 minio_path
     provider: str = "unsplash"  # 图片源: "pexels" 或 "unsplash"
 
@@ -29,71 +29,70 @@ class ExplorerState(TypedDict):
 """节点"""
 
 
-def extract_input_node(state: ExplorerState) -> dict:
+def extract_input_node(state: exploreState) -> dict:
     """从 messages 中提取用户输入"""
-    input_text = state["messages"][0].content if state.get("messages") else ""
+    input_text = state["messages"][-1].content if state.get("messages") else input_text
     return {"input_text": input_text}
 
 
 class SearchQuery(BaseModel):
     """搜索关键词"""
 
-    query: str = Field(description="用于搜索灵感图片的英文关键词，简洁有力")
+    query: str = Field(description="用于搜索灵感图片的英文关键词,简洁有力")
 
 
-# TODO 要考虑搜索图片失败或者图片不存在的情况， 搜索不到 需要调整搜索词或者拆分搜索词，最终失败的话调用mood board生成工具生成， 保证绝对有图片
-async def generate_query_node(state: ExplorerState) -> dict:
-    """使用 LLM 分析用户输入，生成搜索关键词"""
+# TODO 要考虑搜索图片失败或者图片不存在的情况, 搜索不到 需要调整搜索词或者拆分搜索词,最终失败的话调用mood board生成工具生成, 保证绝对有图片
+async def generate_query_node(state: exploreState) -> dict:
+    """使用 LLM 分析用户输入,生成搜索关键词"""
     input_text = state["input_text"]
-    logger.info(f"[Explorer] 用户输入: {input_text}")
-    llm = build_llm()
+    logger.info(f"[explore] 用户输入: {input_text}")
 
-    structured_llm = llm.with_structured_output(SearchQuery)
+    structured_llm = qwen_plus_llm.with_structured_output(SearchQuery)
 
     messages = [
-        SystemMessage(content="""你是一个专业的服装设计师助手。
-根据用户输入，生成一个英文搜索关键词，用于在图片库中搜索服装设计灵感图片（moodboard）。
+        SystemMessage(content="""你是专业服装设计师助手.
+根据用户中文需求,生成适合时尚灵感图(moodboard)图库搜索的英文关键词短句.
 
-要求：
-1. 使用英文，简洁有力
-2. 适合搜索高质量的设计灵感图片
+严格输出规则:
+1. 必须返回标准JSON对象,**禁止输出任何额外文字,解释,思考,前言后语**;
+2. JSON 只包含一个字段 "query",值为简洁英文搜索词;
+3. 关键词简洁,适配高清时尚素材搜索;
 
-例如：
-用户输入："夏季连衣裙，清新风格"
-输出：summer dress fresh style"""),
+输出格式示例(仅允许输出如下JSON,不要加别的内容):
+{"query": "summer dress fresh style"}"""),
         HumanMessage(content=input_text),
     ]
-
     result = structured_llm.invoke(messages)
-    logger.info(f"[Explorer] LLM 生成的搜索关键词: {result.query}")
-    return {"search_query": result.query}
+    search_query = result.query
+    logger.info(f"[explore] LLM 生成的搜索关键词: {search_query}")
+    return {"search_query": search_query}
 
 
-async def search_and_upload_node(state: ExplorerState, config: RunnableConfig) -> dict:
+async def search_and_upload_node(state: exploreState, config: RunnableConfig) -> dict:
     """使用搜索关键词获取图片并上传到 minio"""
     query = state.get("search_query", "")
     user_id = state.get("user_id", "agent")
     provider = state.get("provider", "unsplash")
 
     try:
-        results = await explor_tool.ainvoke({"query": query, "per_page": 4, "user_id": user_id, "method": provider}, config=config)
+        results = await explore_tool.ainvoke({"query": query, "per_page": 4, "user_id": user_id, "method": provider}, config=config)
     except Exception as e:
-        logger.error(f"[Explorer] 搜索失败 '{query}': {e}")
+        logger.error(f"[explore] 搜索失败 '{query}': {e}")
         results = []
 
     return {"image_results": results}
 
 
-def summarize_node(state: ExplorerState) -> dict:
+def summarize_node(state: exploreState) -> dict:
     """汇总结果"""
     input_text = state.get("input_text", "")
     query = state.get("search_query", "")
     results = state.get("image_results", [])
 
-    result_text = f"【灵感探索 Moodboard】\n\n"
-    result_text += f"基于您的需求：「{input_text}」\n"
-    result_text += f"搜索关键词：{query}\n\n"
-    result_text += f"已为您找到 {len(results)} 张灵感图片：\n"
+    result_text = f"[灵感探索 Moodboard]\n\n"
+    result_text += f"基于您的需求: {input_text}\n\n"
+    result_text += f"搜索关键词:{query}\n\n"
+    result_text += f"已为您找到 {len(results)} 张灵感图片:\n"
 
     for i, item in enumerate(results, 1):
         result_text += f"  {i}. 原图: {item.get('image_url', '')}\n"
@@ -105,9 +104,9 @@ def summarize_node(state: ExplorerState) -> dict:
 """构建图"""
 
 
-def build_explorer_graph():
+def build_explore_graph():
     """构建灵感探索图"""
-    workflow = StateGraph(ExplorerState)
+    workflow = StateGraph(exploreState)
 
     workflow.add_node("extract_input", extract_input_node)
     workflow.add_node("generate_query", generate_query_node)
@@ -126,10 +125,10 @@ def build_explorer_graph():
 if __name__ == "__main__":
 
     async def test():
-        graph = build_explorer_graph()
+        graph = build_explore_graph()
         result = await graph.ainvoke(
             {
-                "messages": [HumanMessage(content="夏季连衣裙，清新自然风格")],
+                "messages": [HumanMessage(content="夏季连衣裙,清新自然风格")],
                 "provider": "unsplash",
             }
         )

app/service/fashion_agent/graph_node/explore_graph/tools.py

@@ -15,7 +15,7 @@ class SearchInput(BaseModel):
 
 
 @tool(args_schema=SearchInput)
-async def explor_tool(
+async def explore_tool(
     query: str, per_page: int = 4, user_id: str = "agent", method: str = "unsplash", config: RunnableConfig = None
 ) -> list[dict]:
     """Search for fashion inspiration images on Unsplash and upload to minio. Returns a list of dicts with image_url and minio_path."""

app/service/fashion_agent/graph_node/logo_graph/graph.py

@@ -7,7 +7,7 @@ from langgraph.graph import END, START, StateGraph
 from langgraph.graph.message import add_messages
 from pydantic import BaseModel, Field
 from app.service.fashion_agent.graph_node.logo_graph.tools import generate_logo_tool
-from app.service.fashion_agent.init_llm import qwen_plus_llm
+from app.service.fashion_agent.init_llm import build_llm, qwen_plus_llm
 
 """初始化 LLM TODO 将 API Key 替换为环境变量或者配置文件中的值，避免在代码中硬编码敏感信息"""
 
@@ -17,7 +17,7 @@ from app.service.fashion_agent.init_llm import qwen_plus_llm
 
 class LogoState(TypedDict):
     messages: Required[Annotated[list[AnyMessage], add_messages]]
-    input_text: str
+    input_text: str = ""
     user_id: str = "agent"
     role: str = ""
     gender: str = ""
@@ -35,14 +35,14 @@ class LogoState(TypedDict):
 
 # 定义输出结构
 class LogoPrompt(BaseModel):
-    """生成的 Logo 图像提示词"""
+    """logo image generation diagram prompt words"""
 
-    prompts: list[str] = Field(description="用于生成 Logo 的详细提示词")
+    prompts: list[str] = Field(description="Array of prompt words, simple English words")
 
 
 def extract_input_node(state: LogoState) -> dict:
     """从 messages 中提取用户输入"""
-    input_text = state["messages"][0].content if state.get("messages") else ""
+    input_text = state["messages"][-1].content if state.get("messages") else state["input_text"]
     return {"input_text": input_text}
 
 
@@ -51,18 +51,19 @@ def generate_logo_prompt_node(state: LogoState) -> dict:
     structured_llm = qwen_plus_llm.with_structured_output(LogoPrompt)
 
     messages = [
-        SystemMessage(content="""从用户输入中提取核心主题词，只输出一个简单的英文单词。
+        SystemMessage(content="""从用户输入中提取核心主题词，用一个简单的英文单词作为 prompts 字段的值。
     例如：
-    - "我想要一个猫咪图案" -> "cat"
-    - "设计一个花朵" -> "flower"
-    - "可爱的狗" -> "dog"
-    只输出单词，不要其他内容。"""),
+    - "我想要一个猫咪图案" -> {"prompts": ["cat"]}
+    - "设计一个花朵" -> {"prompts": ["flower"]}
+    - "可爱的狗" -> {"prompts": ["dog"]}
+    请严格按照 JSON 格式输出，包含 prompts 字段。
+    """),
         HumanMessage(content=state["input_text"]),
     ]
 
     result = structured_llm.invoke(messages)
     prompts = result.prompts
-
+    print(result)
     return {
         "logo_prompts": prompts,
     }
@@ -139,14 +140,14 @@ async def main(test_input, user_id="agent", need_prompt_generation=True):
 
 
 if __name__ == "__main__":
-    # 测试示例 1: 需要 prompt 生成（默认）- 简单关键词输入
-    test_input = "我想要一个金毛图案"
-    result = asyncio.run(main(test_input, need_prompt_generation=True))
-    print("=== 需要 prompt 生成 ===")
-    print(f"Result: {result}")
 
-    # 测试示例 2: 直接使用用户提供的 prompt
-    user_prompt = "golden retriever"
-    result = asyncio.run(main(user_prompt, need_prompt_generation=False))
-    print("\n=== 直接使用 prompt ===")
-    print(f"Result: {result}")
+    async def test():
+        graph = build_logo_graph()
+        result = await graph.ainvoke(
+            {
+                "messages": [HumanMessage(content="我想要一个金毛图案")],
+            }
+        )
+        print(result["messages"][-1].content)
+
+    asyncio.run(test())

app/service/fashion_agent/graph_node/node_tools/generate_image.py

@@ -4,13 +4,14 @@ import httpx
 async def generate_image(
     bucket_name="fida-public-bucket",
     object_name=f"furniture/sketches/123456.png",
+    input_images=[],
     prompt="Generate a modern minimalist dining chair made of light "
     "oak wood and white leather, with slim metal legs, photographed "
     "in a bright Scandinavian living room with natural sunlight, high detail, "
     "8k resolution.",
 ):
     request_data = {
-        "input_image_paths": [],
+        "input_image_paths": input_images,
         "prompt": prompt,
         "bucket_name": bucket_name,
         "object_name": object_name,

app/service/fashion_agent/graph_node/node_tools/pexels_search.py

@@ -63,10 +63,10 @@ async def search_photos(query: str, per_page: int = 4, user_id: str = "agent") -
             # 上传到 minio（使用线程池避免阻塞事件循环）
             file_name = f"{uuid7()}.jpg"
             loop = asyncio.get_event_loop()
-            minio_url = await loop.run_in_executor(executor, upload_SDXL_image, image, user_id, "explorer", file_name)
+            minio_url = await loop.run_in_executor(executor, upload_SDXL_image, image, user_id, "explore", file_name)
             results.append({"image_url": image_url, "minio_path": minio_url})
-            logger.info(f"[Explorer] 上传成功: {minio_url}")
+            logger.info(f"[explore] 上传成功: {minio_url}")
         except Exception as e:
-            logger.error(f"[Explorer] 上传失败: {e}")
+            logger.error(f"[explore] 上传失败: {e}")
 
     return results

app/service/fashion_agent/graph_node/node_tools/unsplash_search.py

@@ -63,11 +63,11 @@ async def get_random_photos(query: str, count: int = 4, user_id: str = "agent")
             # 上传到 minio（使用线程池避免阻塞事件循环）
             file_name = f"{uuid7()}.jpg"
             loop = asyncio.get_event_loop()
-            minio_url = await loop.run_in_executor(executor, upload_SDXL_image, image, user_id, "explorer", file_name)
+            minio_url = await loop.run_in_executor(executor, upload_SDXL_image, image, user_id, "explore", file_name)
             results.append({"image_url": image_url, "minio_path": minio_url})
-            logger.info(f"[Explorer] 上传成功: {minio_url}")
+            logger.info(f"[explore] 上传成功: {minio_url}")
         except Exception as e:
-            logger.error(f"[Explorer] 上传失败: {e}")
+            logger.error(f"[explore] 上传失败: {e}")
 
     return results
 

app/service/fashion_agent/graph_node/print_graph/graph.py

@@ -40,7 +40,7 @@ class PrintPrompt(BaseModel):
 
 def extract_input_node(state: PrintState) -> dict:
     """从 messages 中提取用户输入"""
-    input_text = state["messages"][0].content if state.get("messages") else ""
+    input_text = state["messages"][-1].content if state.get("messages") else ""
     return {"input_text": input_text}
 
 
@@ -49,13 +49,13 @@ def generate_print_prompt_node(state: PrintState) -> dict:
     structured_llm = qwen_plus_llm.with_structured_output(PrintPrompt)
 
     messages = [
-        SystemMessage(content=f"""你是一个专业的印花图案设计师。
-    请根据用户输入，生成用于AI图像生成的印花图案提示词。
+        SystemMessage(content=f"""你是一个专业的印花图案设计师.
+    请根据用户输入,生成用于AI图像生成的印花图案提示词.
 
-    要求：
-    1. 提示词应该详细描述印花图案的样式、元素、颜色、布局
+    要求:
+    1. 提示词应该详细描述印花图案的样式,元素,颜色,布局
     2. 提示词应该适合用于 Stable Diffusion 图像生成模型
-    3. 提示词应该使用英文，因为图像生成模型对英文理解更好
+    3. 提示词应该使用英文,因为图像生成模型对英文理解更好
     4. 提示词数量为 {state.get("print_num", 1)}
     """),
         HumanMessage(content=state["input_text"]),
@@ -73,17 +73,19 @@ def generate_print_prompt_node(state: PrintState) -> dict:
 
 
 async def generate_print_img_node(state: PrintState) -> dict:
-    """根据生成的提示词，生成印花图案"""
-    # 如果 print_prompts 为空，使用 input_text 作为 prompt
+    """根据生成的提示词,生成印花图案"""
+    # 如果 print_prompts 为空,使用 input_text 作为 prompt
     if state.get("print_need_prompt_generation", False):
         prompts = state["print_prompts"] if state["print_prompts"] else [state["input_text"]]
     else:
         input_text = state.get("input_text", "")
         prompts = [input_text]
 
+    input_images = state.get("input_images", [])
+
     print_img_urls = []
     for prompt in prompts:
-        image_url = await generate_print_tool.ainvoke({"prompt": prompt})
+        image_url = await generate_print_tool.ainvoke({"prompt": prompt, "input_images": input_images})
         print_img_urls.append(image_url)
         logger.info(f"[Print Graph] Generated print image URL: {image_url}")
 
@@ -94,7 +96,7 @@ async def generate_print_img_node(state: PrintState) -> dict:
 
 
 def should_generate_prompt(state: PrintState) -> str:
-    """条件分支：判断是否需要生成 prompt"""
+    """条件分支:判断是否需要生成 prompt"""
 
     logger.info(
         f"[Print Graph] should_generate_prompt: print_need_prompt_generation={state.get('print_need_prompt_generation')}, print_prompts={state.get('print_prompts')}"
@@ -106,7 +108,6 @@ def should_generate_prompt(state: PrintState) -> str:
 
 
 def build_print_graph():
-
     workflow = StateGraph(PrintState)
     workflow.add_node("extract_input", extract_input_node)
     workflow.add_node("gen_prompt", generate_print_prompt_node)
@@ -145,8 +146,8 @@ async def main(test_input, print_need_prompt_generation=True):
 
 
 if __name__ == "__main__":
-    # 测试示例 1: 需要 prompt 生成（默认）
-    test_input = "我想要一个优雅的花卉印花，适合用于连衣裙，颜色以粉色和白色为主"
+    # 测试示例 1: 需要 prompt 生成(默认)
+    test_input = "我想要一个优雅的花卉印花,适合用于连衣裙,颜色以粉色和白色为主"
     result = asyncio.run(main(test_input, print_need_prompt_generation=True))
     print("=== 需要 prompt 生成 ===")
     print(f"Result: {result}")

app/service/fashion_agent/graph_node/print_graph/tools.py

@@ -11,15 +11,16 @@ class GenerateImageToolInput(BaseModel):
     """Input schema for the Generate Image Tool."""
 
     prompt: str = Field(description="Description of the desired image, e.g., 'A cozy living room with warm lighting and natural textures.'")
+    input_images: list[str] = Field(default=[], description="Input images for the generation.")
 
 
 @tool(args_schema=GenerateImageToolInput)
-async def generate_print_tool(prompt: str) -> str:
+async def generate_print_tool(prompt: str, input_images: list[str]) -> str:
     """Generate an image based on the provided prompt."""
 
     bucket_name = "aida-users"
     object_name = f"agent_generate_print/{uuid7()}.png"
-    image_url = await generate_image(prompt=prompt, bucket_name=bucket_name, object_name=object_name)
+    image_url = await generate_image(prompt=prompt, bucket_name=bucket_name, object_name=object_name, input_images=input_images)
     return [image_url]
 
 

app/service/fashion_agent/graph_node/sketch_graph/graph.py

@@ -42,7 +42,7 @@ class SketchPrompt(BaseModel):
 
 def extract_input_node(state: SketchState) -> dict:
     """从 messages 中提取用户输入"""
-    input_text = state["messages"][0].content if state.get("messages") else ""
+    input_text = state["messages"][-1].content if state.get("messages") else ""
     return {"input_text": input_text}
 
 
@@ -80,15 +80,16 @@ async def generate_sketch_img_node(state: SketchState) -> dict:
     """根据生成的提示词，生成服装草图"""
     # 如果 sketch_need_prompt_generation=False 且 sketch_prompts 为空，使用模板生成 prompt
     if not state.get("sketch_need_prompt_generation", False) and not state.get("sketch_prompts"):
-
         input_text = state.get("input_text", "")
         prompts = [build_sketch_template_prompt(input_text)]
     else:
         prompts = state["sketch_prompts"] if state["sketch_prompts"] else [state["input_text"]]
 
+    input_images = state.get("input_images", [])
+
     sketch_img_urls = []
     for prompt in prompts:
-        image_url = await generate_sketch_tool.ainvoke({"prompt": prompt})
+        image_url = await generate_sketch_tool.ainvoke({"prompt": prompt, "input_images": input_images})
         sketch_img_urls.append(image_url)
 
     return {"sketch_img_urls": sketch_img_urls}
@@ -107,33 +108,26 @@ def should_generate_prompt(state: SketchState) -> str:
 
 def build_sketch_graph():
     workflow = StateGraph(SketchState)
+    workflow.add_node("extract_input", extract_input_node)
+    workflow.add_node("gen_prompt", generate_sketch_prompt_node)
     workflow.add_node("gen_sketch", generate_sketch_img_node)
-    workflow.add_edge(START, "gen_sketch")
+
+    # 添加边
+    workflow.add_edge(START, "extract_input")
+    workflow.add_conditional_edges(
+        "extract_input",
+        should_generate_prompt,
+        {
+            "gen_prompt": "gen_prompt",
+            "gen_sketch": "gen_sketch",
+        },
+    )
+    workflow.add_edge("gen_prompt", "gen_sketch")
     workflow.add_edge("gen_sketch", END)
+
     graph = workflow.compile()
     return graph
 
-    # workflow = StateGraph(SketchState)
-    # workflow.add_node("extract_input", extract_input_node)
-    # workflow.add_node("gen_prompt", generate_sketch_prompt_node)
-    # workflow.add_node("gen_sketch", generate_sketch_img_node)
-
-    # # 添加边
-    # workflow.add_edge(START, "extract_input")
-    # workflow.add_conditional_edges(
-    #     "extract_input",
-    #     should_generate_prompt,
-    #     {
-    #         "gen_prompt": "gen_prompt",
-    #         "gen_sketch": "gen_sketch",
-    #     },
-    # )
-    # workflow.add_edge("gen_prompt", "gen_sketch")
-    # workflow.add_edge("gen_sketch", END)
-
-    # graph = workflow.compile()
-    # return graph
-
 
 def build_sketch_template_prompt(input_text: str) -> str:
     """构建 sketch prompt 模板"""

app/service/fashion_agent/graph_node/sketch_graph/tools.py

@@ -11,15 +11,16 @@ class GenerateImageToolInput(BaseModel):
     """Input schema for the Generate Image Tool."""
 
     prompt: str = Field(description="Description of the desired image, e.g., 'A cozy living room with warm lighting and natural textures.'")
+    input_images: list[str] = Field(default=[], description="Input images for the generation.")
 
 
 @tool(args_schema=GenerateImageToolInput)
-async def generate_sketch_tool(prompt: str) -> str:
+async def generate_sketch_tool(prompt: str, input_images: list[str]) -> str:
     """Generate an image based on the provided prompt."""
 
     bucket_name = "fida-public-bucket"
     object_name = f"test/{uuid7()}.png"
-    image_url = await generate_image(prompt=prompt, bucket_name=bucket_name, object_name=object_name)
+    image_url = await generate_image(prompt=prompt, bucket_name=bucket_name, object_name=object_name, input_images=input_images)
     return [image_url]
 
 

app/service/fashion_agent/main_agent.py

@@ -1,16 +1,11 @@
-import sys
-from pathlib import Path
 from typing import Annotated, Required, TypedDict
-from deepagents import CompiledSubAgent, create_deep_agent
 from langchain.agents import create_agent
-from langchain.tools import tool
-from langchain_core.messages import AnyMessage, HumanMessage
-from langchain_qwq import ChatQwen
+from langchain_core.messages import AnyMessage
 from langgraph.graph import END, START, StateGraph
 from langgraph.graph.message import add_messages
+
 from app.service.fashion_agent.graph_node.design_graph.graph import build_design_graph
-from app.service.fashion_agent.graph_node.design_graph.tools import design_tool
-from app.service.fashion_agent.graph_node.explorer_graph.tools import explor_tool
+from app.service.fashion_agent.graph_node.explore_graph.tools import explore_tool
 from app.service.fashion_agent.graph_node.logo_graph.graph import build_logo_graph
 from app.service.fashion_agent.graph_node.logo_graph.tools import generate_logo_tool
 from app.service.fashion_agent.graph_node.print_graph.graph import build_print_graph
@@ -18,7 +13,7 @@ from app.service.fashion_agent.graph_node.print_graph.tools import generate_prin
 from app.service.fashion_agent.graph_node.sketch_graph.graph import build_sketch_graph
 from app.service.fashion_agent.graph_node.sketch_graph.tools import generate_sketch_tool
 from app.service.fashion_agent.graph_node.trending_graph.trending_graph import build_trending_graph
-from app.service.fashion_agent.graph_node.explorer_graph.graph import build_explorer_graph
+from app.service.fashion_agent.graph_node.explore_graph.graph import build_explore_graph
 from app.service.fashion_agent.init_llm import build_llm
 
 print_graph = build_print_graph()
@@ -26,13 +21,16 @@ logo_graph = build_logo_graph()
 sketch_graph = build_sketch_graph()
 design_graph = build_design_graph()
 trending_graph = build_trending_graph()
-explorer_graph = build_explorer_graph()
+explore_graph = build_explore_graph()
 
 
 class MainState(TypedDict):
     # 消息
     messages: Required[Annotated[list[AnyMessage], add_messages]]
 
+    # 上传图片
+    input_images: list[str] = []
+
     # 模块控制
     call_design: bool = False
     call_print: bool = False
@@ -40,7 +38,7 @@ class MainState(TypedDict):
     call_sketch: bool = False
     call_design: bool = False
     call_trending: bool = False
-    call_explor: bool = False
+    call_explore: bool = False
 
     # design参数
     design_request_data: dict = {}
@@ -58,7 +56,7 @@ class MainState(TypedDict):
     print_img_urls: list[str] = []
 
 
-tools = [explor_tool, generate_logo_tool, generate_print_tool, generate_sketch_tool]
+tools = [explore_tool, generate_logo_tool, generate_print_tool, generate_sketch_tool]
 
 
 def route_node(state: MainState) -> str:
@@ -73,12 +71,12 @@ def route_node(state: MainState) -> str:
         return "direct_design"
     if state.get("call_trending"):
         return "direct_trending"
-    if state.get("call_explor"):
-        return "direct_explor"
+    if state.get("call_explore"):
+        return "direct_explore"
     return "llm_agent"
 
 
-def build_main_graph(enable_thinking: bool = False):
+async def build_main_graph(enable_thinking: bool = False, checkpointer=None):
     llm = build_llm(enable_thinking=enable_thinking)
     chat_agent = create_agent(
         model=llm, tools=tools, state_schema=MainState, system_prompt="你是一个专业的服装设计助手。根据用户需求，调用合适的工具完成任务."
@@ -94,7 +92,7 @@ def build_main_graph(enable_thinking: bool = False):
     workflow.add_node("direct_sketch", sketch_graph)
     workflow.add_node("direct_design", design_graph)
     workflow.add_node("direct_trending", trending_graph)
-    workflow.add_node("direct_explor", explorer_graph)
+    workflow.add_node("direct_explore", explore_graph)
 
     # 条件分支
     workflow.add_conditional_edges(
@@ -107,7 +105,7 @@ def build_main_graph(enable_thinking: bool = False):
             "direct_sketch": "direct_sketch",
             "direct_design": "direct_design",
             "direct_trending": "direct_trending",
-            "direct_explor": "direct_explor",
+            "direct_explore": "direct_explore",
         },
     )
 
@@ -118,27 +116,7 @@ def build_main_graph(enable_thinking: bool = False):
     workflow.add_edge("direct_sketch", END)
     workflow.add_edge("direct_design", END)
     workflow.add_edge("direct_trending", END)
-    workflow.add_edge("direct_explor", END)
+    workflow.add_edge("direct_explore", END)
 
-    return workflow.compile()
-
-
-agent = build_main_graph()
-
-
-if __name__ == "__main__":
-    import asyncio
-
-    async def test_direct():
-        # 直接调用 sketch，跳过 LLM
-        result = await agent.ainvoke(
-            {
-                "messages": [HumanMessage(content="我想设计衬衫，带有猫咪的图案")],
-                "call_sketch": True,
-                "sketch_need_prompt_generation": False,
-            }
-        )
-        print("=== 直接调用 sketch ===")
-        print(result["messages"][-1].content)
-
-    asyncio.run(test_direct())
+    graph = workflow.compile(checkpointer=checkpointer)
+    return graph

app/service/fashion_agent/service.py

@@ -1,13 +1,12 @@
 import json
 import logging
-import sys
-from pathlib import Path
 from langgraph.stream import ProtocolEvent, StreamChannel, StreamTransformer
 from app.service.fashion_agent.main_agent import build_main_graph
 from langgraph.prebuilt import ToolCallTransformer
-from typing import AsyncGenerator, TypedDict
+from typing import AsyncGenerator
 from langchain_core.messages import HumanMessage, ToolMessage
 from app.schemas.fashion_agent import FashionAgentRequest
+from langgraph.checkpoint.postgres.aio import AsyncPostgresSaver
 
 logger = logging.getLogger()
 
@@ -33,73 +32,83 @@ class FashionAgentService:
     async def run_stream(self, request: FashionAgentRequest) -> AsyncGenerator[str, None]:
         """流式运行 agent - 使用 v3 projections"""
 
-        config = {"configurable": {"user_id": request.user_id}}
+        config = {"configurable": {"thread_id": request.thread_id, "user_id": request.user_id}}
 
-        agent = build_main_graph(enable_thinking=request.enable_thinking)
-        state = {
-            "messages": [HumanMessage(content=request.message)],
-            "call_print": request.call_print,
-            "call_logo": request.call_logo,
-            "call_sketch": request.call_sketch,
-            "call_design": request.call_design,
-            "call_trending": request.call_trending,
-            "call_explor": request.call_explor,
-            "print_need_prompt_generation": request.print_need_prompt_generation,
-            "sketch_need_prompt_generation": request.sketch_need_prompt_generation,
-            "design_request_data": request.design_request_data,
-        }
+        async with AsyncPostgresSaver.from_conn_string("postgresql://postgres:Aidlab123123!@20.1.1.43:15432/myapp_prod") as checkpointer:
+            await checkpointer.setup()
+            agent = await build_main_graph(enable_thinking=request.enable_thinking, checkpointer=checkpointer)
 
-        stream = await agent.astream_events(state, config=config, version="v3", transformers=[ToolCallTransformer, CustomTransformer])
+            state = {
+                "messages": [HumanMessage(content=request.message)],
+                "input_images": request.input_images,
+                "call_print": request.call_print,
+                "call_logo": request.call_logo,
+                "call_sketch": request.call_sketch,
+                "call_design": request.call_design,
+                "call_trending": request.call_trending,
+                "call_explore": request.call_explore,
+                "print_need_prompt_generation": request.print_need_prompt_generation,
+                "sketch_need_prompt_generation": request.sketch_need_prompt_generation,
+                "design_request_data": request.design_request_data,
+            }
 
-        tool_names = {}
-        filter_tool_name = ["design_tool"]
-        async for event in stream:
-            if event["method"] == "tools":
-                data = event["params"]["data"]
-                tool_call_id = data.get("tool_call_id")
+            stream = await agent.astream_events(state, config=config, version="v3", transformers=[ToolCallTransformer, CustomTransformer])
 
-                # 记录 tool_name
-                if data.get("event") == "tool-started":
-                    tool_names[tool_call_id] = data.get("tool_name")
+            tool_names = {}
+            filter_tool_name = ["design_tool"]
+            async for event in stream:
+                if event["method"] == "tools":
+                    data = event["params"]["data"]
+                    tool_call_id = data.get("tool_call_id")
 
-                # 通过 ID 查找 tool_name
-                elif data.get("event") == "tool-finished":
-                    tool_name = tool_names.get(tool_call_id, "unknown")
+                    # 记录 tool_name
+                    if data.get("event") == "tool-started":
+                        tool_names[tool_call_id] = data.get("tool_name")
 
-                    if tool_name in filter_tool_name:
+                    # 通过 ID 查找 tool_name
+                    elif data.get("event") == "tool-finished":
+                        tool_name = tool_names.get(tool_call_id, "unknown")
+
+                        if tool_name in filter_tool_name:
+                            continue
+
+                        data["tool_name"] = tool_name
+
+                        if isinstance(data["output"], ToolMessage):
+                            data["output"] = json.loads(data["output"].content)
+
+                    response_event = {"event_type": "tool", "data": data}
+                    yield f"data: {json.dumps(response_event, ensure_ascii=False)}\n\n"
+
+                elif event["method"] == "custom":
+                    data = event["params"]["data"]
+                    response_event = {"event_type": "tool", "data": data}
+                    yield f"data: {json.dumps(response_event, ensure_ascii=False)}\n\n"
+
+                elif event["method"] == "messages":
+                    event_data = event["params"]["data"]
+                    data = event_data[0] if len(event_data) > 0 else {}
+                    # 提取元数据 (如果有的话)
+                    metadata = event_data[1] if len(event_data) > 1 else {}
+                    if not isinstance(data, dict):
+                        continue
+                    if metadata.get("langgraph_node") in {"gen_prompt", "generate_query"}:
                         continue
 
-                    data["tool_name"] = tool_name
+                    ev = data.get("event")
 
-                    if isinstance(data["output"], ToolMessage):
-                        data["output"] = json.loads(data["output"].content)
-                response_event = {"event_type": "tool", "data": data}
-                yield f"data: {json.dumps(response_event, ensure_ascii=False)}\n\n"
+                    if ev == "content-block-delta":
+                        block = data.get("delta") or {}
+                        if block.get("type") in ("text-delta", "reasoning-delta"):
+                            response_event = {"event_type": "messages", "data": {"event": ev} | block}
+                            yield f"data: {json.dumps(response_event, ensure_ascii=False)}\n\n"
 
-            elif event["method"] == "custom":
-                data = event["params"]["data"]
-                response_event = {"event_type": "tool", "data": data}
-                yield f"data: {json.dumps(response_event, ensure_ascii=False)}\n\n"
+                    elif ev in ("message-start", "content-block-start", "content-block-finish", "message-finish"):
+                        response_event = {"event_type": "messages", "data": {"event": ev} | data}
+                        yield f"data: {json.dumps(response_event, ensure_ascii=False)}\n\n"
 
-            elif event["method"] == "messages":
-                data = event["params"]["data"][0]
-                if not isinstance(data, dict):
-                    continue
-                if data.get("event") != "content-block-delta":
-                    continue
-                block = data.get("delta") or {}
-
-                if block.get("type") == "text-delta":
-                    response_event = {"event_type": "messages", "data": {"event": data["event"]} | block}
-                    yield f"data: {json.dumps(response_event, ensure_ascii=False)}\n\n"
-                elif block.get("type") == "reasoning-delta":
-                    response_event = {"event_type": "messages", "data": {"event": data["event"]} | block}
-                    yield f"data: {json.dumps(response_event, ensure_ascii=False)}\n\n"
-                else:
-                    pass
-                    # print(f"----------------{event}")
-        response_event = {"event_type": "done"}
-        yield f"data: {response_event}"
+            response_event = {"event_type": "done"}
+            yield f"data: {response_event}"
 
 
 if __name__ == "__main__":
@@ -117,13 +126,15 @@ if __name__ == "__main__":
         print("测试流式输出")
         print("=" * 50)
         request = FashionAgentRequest(
-            message="生成一张草莓图案",
-            call_print=True,
+            thread_id="zhh",
+            message="落日",
+            # call_print=True,
+            # input_images=["test/53d38bd5-f77b-4034-ada2-45f1e2ebe00c.png"],
             # print_need_prompt_generation=False,
             # call_sketch=True,
             # sketch_need_prompt_generation=False,
             # call_logo=True,
-            # call_explor=True,
+            call_explore=True,
             # call_design=True,
             # design_request_data=request_data,
         )