aidlab/FiDA_Python
7
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat 接入report

Browse Source
This commit is contained in:
zcr
2026-03-03 17:33:51 +08:00
parent 1ecb02d706
commit 1ade907828
23 changed files with 4079 additions and 516 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
.gitignore vendored
View File

@@ -147,3 +147,4 @@ app/logs/*
*.json
*.env*
config.backup.py
*.md

45
config.yaml
View File

@@ -16,20 +16,41 @@ agents:
designer:
prompt_template: |
你是一位资深的家具设计师。你的职责是：
1. 从用户的模糊描述中提取或补充具体的设计参数（尺寸、材质、人体工学数据）。
2. 如果用户想画图，不要直接画，而是先描述清楚细节，然后让 Visualizer 去画。
请以专业的口吻回复。
你是一位资深的家具设计师，经验丰富、审美一流、沟通温暖且高效。
你的核心目标：快速理解用户想法，并用最合适的方式推进设计。
你可以：
1. 用户描述模糊时，可以自然地询问或给出建议，但**绝不强迫补充**尺寸、材质、人体工学等细节，除非用户自己关心或需要明确这些参数。
2. 如果用户提到想看图、想出效果图、想画草图、想渲染等，**直接同意并推动**：
- 用一句话确认或赞美用户的想法
- 主动说“我这就帮你把当前设计转给视觉专家生成效果图/草图”
- 然后让 Visualizer 节点去处理（不需要你先写一大段细节描述）
3. 回复时像和懂设计的客户聊天一样：专业、亲切、有创意，偶尔带点热情或幽默，但始终围绕家具设计。
永远不要用“必须”“请先描述清楚”“按照流程”等强硬的流程化语言。
visualizer:
prompt_template: |
你是视觉专家。你的目标是生成高质量的家具草图。
步骤：
1. 根据上下文，编写一个详细的 Stable Diffusion 风格的英文 Prompt。
2. 必须调用 generate_furniture_sketch 工具来生成图片。
你是专业的家具工业设计视觉专家，擅长将文字描述转化为高质量、清晰、专业的家具设计草图。
注意：如果对话中出现 [SYSTEM_DIRECTIVE] 要求直接绘图，请立即根据已知信息编写 Prompt 并调用 generate_furniture_sketch 工具，不要进行多余的询问。
你的唯一任务：
- 基于当前全部对话上下文（包括用户描述、已有设计要点、风格要求、材质、尺寸、功能等），在内部生成一个详细的英文 Stable Diffusion Prompt。
- 然后**立即调用 generate_furniture 工具**来生成图片。
- **绝对不要**把生成的 Prompt 文本、任何代码块、任何解释、任何思考过程输出给用户。
- 只通过工具调用返回结果，工具执行完成后自然结束。
researcher:
prompt_template: |
你是情报专家，负责检索与整理参考资料并生成报告。
Prompt 内部生成要求（仅供你自己参考，不要输出）：
1. 语言：全程英文
2. 结构：主体描述 + 风格 + 视角 + 细节 + 材质 + 照明 + 质量修饰词
3. 必须包含高质量关键词：highly detailed, sharp focus, professional industrial design sketch, clean line art 或 photorealistic product render
4. 背景：pure white background, studio lighting, no people, no text, no watermark
5. 避免：blurry, deformed, low quality, cartoon, extra limbs, bad anatomy
6. 如果上下文有明确风格、视角，必须强烈体现
7. 长度：80-150 个词左右
现在开始工作：
- 直接分析上下文
- 内部构建 Prompt
- 立即调用 generate_furniture 工具
- 不要输出任何文字

56
logging_env.py Normal file
View File

@@ -0,0 +1,56 @@
import os
from src.core.config import settings
LOGGER_CONFIG_DICT = {
'version': 1,
'disable_existing_loggers': False,
'formatters': {
'simple': {
'format': '%(asctime)s %(filename)s [line:%(lineno)d] %(levelname)s %(message)s',
'datefmt': '%Y-%m-%d %H:%M:%S' # 补充日期格式，日志更易读
}
},
'handlers': {
'console': {
'class': 'logging.StreamHandler',
'level': 'INFO',
'formatter': 'simple',
'stream': 'ext://sys.stdout',
},
'info_file_handler': {
'class': 'logging.handlers.RotatingFileHandler',
'level': 'INFO',
'formatter': 'simple',
'filename': os.path.join(settings.LOGS_PATH, 'info.log'),
'maxBytes': 10485760,
'backupCount': 50,
'encoding': 'utf8',
},
'error_file_handler': {
'class': 'logging.handlers.RotatingFileHandler',
'level': 'ERROR',
'formatter': 'simple',
'filename': os.path.join(settings.LOGS_PATH, 'error.log'),
'maxBytes': 10485760,
'backupCount': 20,
'encoding': 'utf8',
},
'debug_file_handler': {
'class': 'logging.handlers.RotatingFileHandler',
'level': 'DEBUG',
'formatter': 'simple',
'filename': os.path.join(settings.LOGS_PATH, 'debug.log'),
'maxBytes': 10485760,
'backupCount': 50,
'encoding': 'utf8',
},
},
'loggers': {
'my_module': {'level': 'INFO', 'handlers': ['console'], 'propagate': 'no'}
},
'root': {
'level': 'DEBUG',
'handlers': ['error_file_handler', 'info_file_handler', 'debug_file_handler', 'console'],
},
}

6
main.py
View File

@@ -1,8 +1,14 @@
import logging
import uvicorn
from fastapi import FastAPI
from fastapi.middleware.cors import CORSMiddleware
from logging_env import LOGGER_CONFIG_DICT
from src.routers import chat
logging.config.dictConfig(LOGGER_CONFIG_DICT)
app_server = FastAPI(
title="Gemini Furniture Designer API",
description="基于 LangGraph + Gemini 2.0 Flash 的家具设计 Agent 接口",

18
pyproject.toml
View File

@@ -4,20 +4,36 @@ version = "0.1.0"
description = "Add your description here"
requires-python = ">=3.12"
dependencies = [
"crawl4ai>=0.8.0",
"deepagents>=0.4.3",
"fastapi[standard]>=0.128.0",
"gunicorn>=25.0.1",
"image>=1.5.33",
"langchain-community>=0.4.1",
"langchain-core>=1.2.8",
"langchain-google-genai>=4.2.0",
"langgraph>=1.0.7",
"langgraph[postgres]>=1.0.7",
"langgraph-checkpoint-mongodb>=0.3.1",
"minio>=7.2.20",
"modality>=0.1.0",
"motor>=3.7.1",
"playwright>=1.58.0",
"pydantic>=2.12.5",
"pydantic-settings>=2.12.0",
"pymongo[srv]>=4.15.5",
"python-dotenv>=1.2.1",
"tavily-python>=0.7.21",
"uuid>=1.30",
"uvicorn>=0.40.0",
"psycopg[binary]>=3.3.3",
"postgres>=4.0",
"langchain-huggingface>=1.2.0",
"sentence-transformers>=5.2.3",
"rank-bm25>=0.2.2",
"torch>=2.10.0",
"faiss-cpu>=1.13.2",
"terminate>=0.0.9",
"report-generator>=0.1.10",
"dashscope>=1.25.13",
"prompt>=0.4.1",
]

8
src/core/config.py
View File

@@ -17,6 +17,9 @@ class Settings(BaseSettings):
GOOGLE_CLOUD_PROJECT: str = Field(default="", description="")
GOOGLE_CLOUD_LOCATION: str = Field(default="", description="")
# --- google api 配置信息 ---
QWEN_API_KEY: str = Field(default="", description="")
# --- minio 配置信息 ---
MINIO_URL: str = Field(default='', description="")
MINIO_ACCESS: str = Field(default='', description="")
@@ -29,6 +32,11 @@ class Settings(BaseSettings):
MONGODB_HOST: str = Field(default="localhost", description="")
MONGODB_PORT: int = Field(default=27017, description="")
# --- 外部工具api配置信息 ---
TAVILY_API_KEY: str = Field(default="", description="")
LOGS_PATH: str = Field(default="/mnt/data/FiDA/logs", description="")
settings = Settings()
MONGO_URI = f"mongodb://{settings.MONGODB_USERNAME}:{settings.MONGODB_PASSWORD}@{settings.MONGODB_HOST}:{settings.MONGODB_PORT}"

198
src/routers/chat.py
View File

@@ -1,5 +1,8 @@
import logging
import uuid
import json
from typing import AsyncGenerator
from fastapi import APIRouter
from fastapi.responses import StreamingResponse
from src.schemas.chat import ChatRequest, HistoryResponse, HistoryItem
@@ -7,6 +10,7 @@ from src.server.agent.graph import app # 导入已经 compile 好的 graph
from langchain_core.messages import HumanMessage, SystemMessage
router = APIRouter(prefix="/chat", tags=["Furniture Design Chat"])
logger = logging.getLogger(__name__)
@router.post("/stream")
@@ -52,29 +56,28 @@ async def chat_stream(request: ChatRequest):
}
```
"""
logger.debug(f"chat request data: {request}")
source_thread_id = request.thread_id
checkpoint_id = request.checkpoint_id
# 1. 确定目标 thread_id
# 如果是回溯操作，我们生成一个新的 ID，或者由前端传入一个新的 target_thread_id
is_branching = source_thread_id and checkpoint_id
target_thread_id = str(uuid.uuid4())[:8] if is_branching else (source_thread_id or str(uuid.uuid4())[:8])
# 2. 获取配置参数
temp = request.config_params.temperature if request.config_params else 0.7
# 构建基础 Config
# 2. 配置参数
temp = request.config_params.temperature if request.config_params else 0.7
current_config = {
"recursion_limit": 100,
"configurable": {
"thread_id": target_thread_id,
"llm_temperature": temp
"llm_temperature": temp,
"use_report": request.use_report,
}
}
# 3. 处理状态初始化与分支
initial_messages = []
# 如果是全新的对话（没有 source_thread_id），或者明确要求分叉
# 3. 初始化消息 + 系统提示
initial_messages = []
if not source_thread_id or is_branching:
# 如果用户传了标签，构造 SystemMessage 注入上下文
if request.config_params:
cp = request.config_params
system_prompt = (
@@ -86,7 +89,7 @@ async def chat_stream(request: ChatRequest):
)
initial_messages.append(SystemMessage(content=system_prompt))
# 4. 执行分叉逻辑（搬运旧数据）
# 4. 处理分支（从历史 checkpoint 复制状态）
if is_branching:
source_config = {
"configurable": {
@@ -95,80 +98,149 @@ async def chat_stream(request: ChatRequest):
}
}
older_state = await app.aget_state(source_config)
# 将旧消息和我们新定义的 SystemMessage 合并
# update_state 会将这些消息推送到新 thread 的存储中
combined_values = older_state.values.copy()
if initial_messages:
combined_values["messages"] = list(combined_values["messages"]) + initial_messages
combined_values["messages"] = list(combined_values.get("messages", [])) + initial_messages
await app.aupdate_state(current_config, combined_values)
async def event_generator():
# 初始推送状态信息
async def event_generator() -> AsyncGenerator[str, None]:
# 初始事件
yield f"data: {json.dumps({'thread_id': target_thread_id, 'is_branch': is_branching, 'status': 'start'}, ensure_ascii=False)}\n\n"
# 构造本次请求的输入
# 如果是第一次开始，且有 initial_messages，则连同 user message 一起发送
# --- 核心逻辑：构造本次请求的消息列表 ---
new_messages = []
if not source_thread_id and initial_messages:
new_messages.extend(initial_messages)
# 添加用户消息
# 构造输入
new_messages = initial_messages[:] if not source_thread_id else []
new_messages.append(HumanMessage(content=request.message))
# --- 新增：强制绘图指令注入 ---
# if request.force_sketch:
# force_instruction = HumanMessage(
# content="[SYSTEM_DIRECTIVE]: 用户点击了强制生成按钮。请立即根据当前上下文调用 generate_furniture_sketch 工具生成草图，无需确认。"
# )
# new_messages.append(force_instruction)
input_data = {
"messages": new_messages,
"require_suggestion": request.need_suggestion # 初始由前端决定
"require_suggestion": request.need_suggestion,
"use_report": request.use_report,
}
async for event in app.astream(
# 使用 astream_events v2 + stream_subgraphs=True 来捕获 DeepAgents 内部流式事件
async for event in app.astream_events(
input_data,
current_config,
stream_mode="updates"
version="v2",
config=current_config,
stream_subgraphs=True,
):
for node_name, output in event.items():
if "messages" in output:
# 获取最新 state 以获取 checkpoint_id
state = await app.aget_state(current_config)
current_cp_id = state.config["configurable"].get("checkpoint_id")
event_kind = event["event"]
# 获取当前 checkpoint_id（安全方式，避免 KeyError）
latest_state = await app.aget_state(current_config)
configurable = latest_state.config.get("configurable", {})
current_cp_id = configurable.get("checkpoint_id", "") # 如果没有，返回空字符串
# ────────────────────────────────────────────────
# 1. LLM token 流式输出（主图或子图的逐 token）
# ────────────────────────────────────────────────
if event_kind == "on_chat_model_stream":
chunk = event["data"].get("chunk")
if chunk and chunk.content:
node_name = event.get("name", "Unknown")
# 判断是否来自 Researcher 子图
namespace = event.get("parent_ids", []) or event.get("namespace", [])
if any("Researcher" in str(ns) for ns in namespace):
node_name = "Researcher"
# 遍历本次 update 产生的所有消息
for msg in output["messages"]:
payload = {
"node": node_name,
"content": "",
"content": chunk.content,
"is_delta": True,
"checkpoint_id": current_cp_id,
"image_url": None,
"suggestions": []
}
yield f"data: {json.dumps(payload, ensure_ascii=False)}\n\n"
# ────────────────────────────────────────────────
# 2. 自定义事件（report_delta 等）
# ────────────────────────────────────────────────
elif event_kind == "on_custom_event":
custom_data = event["data"]
if isinstance(custom_data, dict):
if custom_data.get("type") == "report_delta":
payload = {
"node": "Researcher",
"content": custom_data.get("delta", ""),
"is_delta": True,
"checkpoint_id": current_cp_id,
"image_url": None,
"suggestions": []
}
yield f"data: {json.dumps(payload, ensure_ascii=False)}\n\n"
# 可选：报告开始/完成/错误等状态提示
elif custom_data.get("type") in ("report_start", "report_complete", "report_error"):
status_msg = {
"report_start": "Start generating reports...",
"report_complete": "Report generation completed",
"report_error": f"Report generation failed: {custom_data.get('message', '')}"
}.get(custom_data["type"], "")
payload = {
"node": "Researcher",
"content": status_msg,
"is_delta": False,
"checkpoint_id": current_cp_id,
"image_url": None,
"suggestions": []
}
yield f"data: {json.dumps(payload, ensure_ascii=False)}\n\n"
# ────────────────────────────────────────────────
# 3. 节点启动 / 工具启动（进度提示）
# ────────────────────────────────────────────────
elif event_kind in {"on_tool_start", "on_tool_end"}:
tool_name = event.get("name", "unknown_tool")
tool_data = event.get("data", {})
tool_input = tool_data.get("input", "")
tool_output = tool_data.get("output", "")
if event_kind == "on_tool_start":
payload = {
"node": tool_name,
"content": tool_input,
"is_delta": False,
"checkpoint_id": current_cp_id,
"image_url": None,
"suggestions": []
}
yield f"data: {json.dumps(payload, ensure_ascii=False)}\n\n"
else:
if tool_name == "generate_furniture" and isinstance(tool_output, str):
payload = {
"node": tool_name,
"content": "Design sketch has been generated for you.", # 给用户友好的文字提示
"image_url": tool_output, # 直接传 URL 给前端显示
"is_delta": False, # 这是一个完整事件，不是增量
"checkpoint_id": current_cp_id,
"suggestions": []
}
# --- 核心改动：提取建议按钮 ---
# 无论是不是 Suggester 节点，只要消息里带了建议就提取
if hasattr(msg, "additional_kwargs") and "suggestions" in msg.additional_kwargs:
payload["suggestions"] = msg.additional_kwargs["suggestions"]
content = msg.content
# 逻辑判断：MinIO 图片处理
if node_name == "Visualizer" and str(content).endswith("png") and "furniture/sketches" in str(content):
payload["image_url"] = content
payload["content"] = "已为您生成设计草图"
else:
payload["content"] = content
# 如果消息既没有文本、也没有图片、也没有建议（比如中间的 ToolCall 消息），则跳过
if not payload["content"] and not payload["image_url"] and not payload["suggestions"]:
continue
yield f"data: {json.dumps(payload, ensure_ascii=False)}\n\n"
elif tool_name == "topic_research":
payload = {
"node": tool_name,
"content": "Visiting...", # 给用户友好的文字提示
"image_url": None, # 直接传 URL 给前端显示
"search_list": tool_output.content,
"is_delta": False, # 这是一个完整事件，不是增量
"checkpoint_id": current_cp_id,
"suggestions": []
}
yield f"data: {json.dumps(payload, ensure_ascii=False)}\n\n"
else:
# 可选：其他工具的通用处理（debug 或显示结果）
if tool_output:
payload = {
"node": tool_name,
"content": f"tool {tool_name} Execution completed：{str(tool_output)[:200]}...", # 截断避免过长
"is_delta": False,
"checkpoint_id": current_cp_id,
"image_url": None,
"suggestions": []
}
yield f"data: {json.dumps(payload, ensure_ascii=False)}\n\n"
# 流结束
yield f"data: {json.dumps({'status': 'end'}, ensure_ascii=False)}\n\n"
return StreamingResponse(event_generator(), media_type="text/event-stream")
@@ -218,7 +290,7 @@ async def get_chat_history(thread_id: str):
}
```
"""
config = {"configurable": {"thread_id": thread_id}}
config = {"configurable": {"thread_id": thread_id}, }
history_data = []
async for state in app.aget_state_history(config):
msg_content = "Initial"

1
src/schemas/chat.py
View File

@@ -15,6 +15,7 @@ class ChatRequest(BaseModel):
checkpoint_id: Optional[str] = Field(None, description="回溯点的ID，用于从历史点开启新对话")
config_params: Optional[AgentConfig] = None
need_suggestion: bool = False
use_report: bool = False # ← 新增：是否使用深度报告
class HistoryItem(BaseModel):

232
src/server/agent/agents.py
View File

@@ -1,36 +1,59 @@
import os
from google.oauth2 import service_account
from pathlib import Path
from typing import AsyncGenerator, Dict, Any
from deepagents import create_deep_agent
from deepagents.backends import FilesystemBackend
from langchain_core.messages import HumanMessage, SystemMessage, ToolMessage, AIMessage
from langchain_core.runnables import RunnableConfig
from langchain_google_genai import ChatGoogleGenerativeAI
from src.server.agent.state import AgentState
from src.server.agent.tools import generate_2025_report_tool, generate_furniture_sketch
from src.server.agent.config_loader import get_agent_prompt
from src.core.config import settings
from src.server.utils.generate_suggestion import generate_chat_suggestions
from langchain_qwq import ChatQwen
creds = service_account.Credentials.from_service_account_file(
settings.GOOGLE_GENAI_USE_VERTEXAI,
scopes=["https://www.googleapis.com/auth/cloud-platform"],
from src.core.config import settings
from src.server.agent.prompt import SYSTEM_PROMPT
from src.server.agent.state import AgentState
from src.server.agent.tools.generate_furniture_sketch import generate_furniture
from src.server.agent.config_loader import get_agent_prompt
from src.server.agent.tools.crawl_tool import crawl4ai_batch
from src.server.agent.tools.research_tool import topic_research
from src.server.agent.tools.structured_retrieval_tool import structured_retrieval
from src.server.agent.tools.terminate_tool import terminate
from src.server.agent.tools.user_persona_tool import manage_user_persona
from src.server.utils.generate_suggestion import generate_chat_suggestions
from test.report.tools.report_generator_tool import report_generator
# 目前這個主程式檔案所在的目錄
MAIN_DIR = Path(__file__).resolve().parent
# 專案根目錄（因為 main.py 跟 tools/ 同級，所以 parent 就是根）
PROJECT_ROOT = MAIN_DIR
model = ChatQwen(
model="qwen3.5-flash",
max_tokens=3_000,
timeout=None,
max_retries=2,
api_key=settings.QWEN_API_KEY)
tools = [manage_user_persona, topic_research, crawl4ai_batch, structured_retrieval, report_generator, terminate]
research_agent = create_deep_agent(
model=model,
tools=tools,
system_prompt=SYSTEM_PROMPT,
backend=FilesystemBackend(
root_dir=str(PROJECT_ROOT / "agent_workspace"),
virtual_mode=False, # 重要：關掉虛擬模式 → 真的寫硬碟
)
)
# 辅助函数：根据配置动态获取 LLM
def get_model(config: RunnableConfig):
# 从 configurable 中获取温度，默认为 0.5 (对应你之前的设置)
# 这个 key 必须与你在 chat_stream 路由里定义的 "llm_temperature" 一致
temp = config["configurable"].get("llm_temperature", 0.5)
return ChatGoogleGenerativeAI(
model="gemini-2.0-flash",
return ChatQwen(
model="qwen3.5-flash",
max_tokens=3_000,
timeout=None,
max_retries=2,
temperature=temp,
credentials=creds,
project=settings.GOOGLE_CLOUD_PROJECT,
location=settings.GOOGLE_CLOUD_LOCATION,
vertexai=True,
api_key=settings.GOOGLE_API_KEY
)
api_key=settings.QWEN_API_KEY)
# --- 1. Designer Agent (设计顾问) ---
@@ -48,33 +71,158 @@ async def designer_node(state: AgentState, config: RunnableConfig):
return {"messages": [response], "require_suggestion": should_suggest}
# --- 2. Researcher Agent (情报专家) ---
async def researcher_node(state: AgentState, config: RunnableConfig):
"""负责调用报告生成工具"""
model = get_model(config)
tools = [generate_2025_report_tool]
llm_with_tools = model.bind_tools(tools)
async def researcher_node(
state: AgentState,
config: RunnableConfig
) -> AsyncGenerator[Dict[str, Any], None]:
use_report = config["configurable"].get("use_report", False)
if not use_report:
yield {
"messages": [AIMessage(
content="深度报告功能未启用，请通过前端按钮触发。",
name="Researcher"
)],
"next": "Supervisor"
}
return
messages = state["messages"]
system_text = get_agent_prompt("researcher")
system_prompt = SystemMessage(content=system_text)
response = await llm_with_tools.ainvoke([system_prompt] + messages)
last_human = next((m for m in reversed(messages) if isinstance(m, HumanMessage)), None)
if response.tool_calls:
tool_call = response.tool_calls[0]
if tool_call["name"] == "generate_2025_report_tool":
# 这里的工具调用如果也是异步的，建议加 await
result = await generate_2025_report_tool.ainvoke(tool_call["args"])
return {"messages": [response, HumanMessage(content=str(result))]}
if not last_human:
yield {
"messages": [AIMessage(
content="深度研究节点：未找到有效的用户问题",
name="Researcher"
)],
"next": "Supervisor"
}
return
return {"messages": [response]}
full_content = ""
current_step = "正在启动深度报告生成..."
# 初始提示
yield {
"messages": [AIMessage(
content="正在启动深度报告生成...",
name="Researcher",
additional_kwargs={
"current_step": current_step,
"streaming": True
}
)]
}
async for event in research_agent.astream_events(
{"messages": messages[-12:]},
version="v2",
config=config,
stream_subgraphs=True
):
event_type = event["event"]
name = event.get("name", "未知")
if event["event"] == "on_custom_event":
custom_data = event["data"]
# 你的 writer 发的是 dict，所以这里 custom_data 就是你写的 {"type": "report_delta", "delta": "..."}
if isinstance(custom_data, dict) and custom_data.get("type") == "report_delta":
delta = custom_data.get("delta", "")
print(delta, end="", flush=True) # 实时打印，不换行
# ────────────── 工具结束事件：重点处理并 yield 输出 ──────────────
if event["event"] in {"on_tool_start", "on_tool_end"}:
tool_name = event.get("name", "未知")
is_start = event["event"] == "on_tool_start"
if is_start:
tool_input = event["data"].get("input", {})
current_step = f"正在執行工具：{tool_name}"
print(f"| {current_step} | {tool_input}")
yield {
"messages": [AIMessage(
content=full_content,
name="Researcher",
additional_kwargs={
"current_step": current_step,
"tool_name": tool_name,
"tool_input": tool_input,
"tool_status": "start",
"streaming": True
}
)]
}
else: # on_tool_end
tool_output = event["data"].get("output", "")
current_step = f"工具 {tool_name} 已完成"
print(f"| {current_step} | {tool_output}")
yield {
"messages": [AIMessage(
content=full_content,
name="Researcher",
additional_kwargs={
"current_step": current_step,
"tool_name": tool_name,
"tool_output": tool_output,
"tool_status": "end",
"streaming": True
}
)]
}
# ────────────── LLM 内容生成（保持原有逻辑） ──────────────
elif event_type == "on_chat_model_stream":
chunk = event["data"]["chunk"].content or ""
if chunk:
print(chunk, end="", flush=True)
full_content += chunk
if "\n" in chunk or len(full_content) % 4 == 0:
yield {
"messages": [AIMessage(
content=full_content,
name="Researcher",
additional_kwargs={
"current_step": current_step,
"streaming": True
}
)]
}
# ────────────── 其他链路事件（可选补充） ──────────────
elif event_type in ("on_chain_start", "on_chain_end"):
status = "开始" if event_type == "on_chain_start" else "完成"
current_step = f"[{status}] {name.upper()}"
yield {
"messages": [AIMessage(
content=full_content,
name="Researcher",
additional_kwargs={
"current_step": current_step,
"streaming": True
}
)]
}
# 最终输出
yield {
"messages": [AIMessage(
content=full_content.strip() or "报告生成完成",
name="Researcher",
additional_kwargs={
"current_step": "报告已完成",
"streaming": False
}
)],
"next": "Suggester"
}
# --- 3. Visualizer Agent (视觉专家) ---
async def visualizer_node(state: AgentState, config: RunnableConfig):
"""负责将自然语言转化为绘图 Prompt 并调用绘图工具"""
model = get_model(config)
tools = [generate_furniture_sketch]
tools = [generate_furniture]
llm_with_tools = model.bind_tools(tools)
messages = state["messages"]
@@ -85,8 +233,8 @@ async def visualizer_node(state: AgentState, config: RunnableConfig):
if response.tool_calls:
tool_call = response.tool_calls[0]
if tool_call["name"] == "generate_furniture_sketch":
img_url = await generate_furniture_sketch.ainvoke(tool_call["args"])
if tool_call["name"] == "generate_furniture":
img_url = await generate_furniture.ainvoke(tool_call["args"])
return {
"messages": [
response,

27
src/server/agent/graph.py
View File

@@ -1,14 +1,12 @@
import os
from typing import Literal
from google.oauth2 import service_account
from langchain_core.messages import AIMessage
from langchain_google_genai import ChatGoogleGenerativeAI
from langchain_core.runnables import RunnableConfig
from langchain_qwq import ChatQwen
from langgraph.graph import StateGraph, END, START
from pydantic import BaseModel
from pymongo import MongoClient
from src.core.config import settings, MONGO_URI
from src.core.config import MONGO_URI, settings
from src.server.agent.state import AgentState
from src.server.agent.agents import designer_node, researcher_node, visualizer_node, suggester_node
from langgraph.checkpoint.mongodb import MongoDBSaver
@@ -21,18 +19,16 @@ class RouteResponse(BaseModel):
next: Literal["Designer", "Researcher", "Visualizer", "Suggester", "FINISH"]
creds = service_account.Credentials.from_service_account_file(
settings.GOOGLE_GENAI_USE_VERTEXAI,
scopes=["https://www.googleapis.com/auth/cloud-platform"],
)
llm_supervisor = ChatGoogleGenerativeAI(
model="gemini-2.0-flash", credentials=creds,
project="aida-461108", location='us-central1', vertexai=True, api_key=settings.GOOGLE_API_KEY
)
llm_supervisor = ChatQwen(
model="qwen3.5-flash",
max_tokens=3_000,
timeout=None,
max_retries=2,
api_key=settings.QWEN_API_KEY)
def supervisor_node(state: AgentState):
def supervisor_node(state: AgentState, config: RunnableConfig):
use_report = config["configurable"].get("use_report", False)
messages = state["messages"]
if not messages:
return {"next": "Suggester"}
@@ -69,7 +65,6 @@ workflow.add_node("Designer", designer_node)
workflow.add_node("Researcher", researcher_node)
workflow.add_node("Visualizer", visualizer_node)
workflow.add_node("Suggester", suggester_node) # 新增节点
workflow.add_edge(START, "Supervisor")
# 修改条件边映射

66
src/server/agent/prompt.py Normal file
View File

@@ -0,0 +1,66 @@
SYSTEM_PROMPT = """
You are "TrendAgent" - a focused, efficient design trend analysis agent.
Your ONLY goal: produce one high-quality Markdown trend report per user request.
TOOL ORDER & DISCIPLINE IS MANDATORY - DO NOT INVENT STEPS
┌───────────────────────────────────────────────────────┐
│ Phase 0 - Context & Persona (必须先完成) │
└───────────────────────────────────────────────────────┘
Rules for Phase 0:
1. ALWAYS start with manage_user_persona(command="get")
2. If STATUS == "INCOMPLETE" or persona missing critical fields (Design Type, Style, Target Audience, Color Preference, etc.):
→ MUST call manage_user_persona(command="ask") to collect missing info
→ After user answers → call manage_user_persona(command="set", ...)
→ Loop until STATUS == "READY"
3. Only when STATUS == "READY" → proceed to Phase 1
4. Never assume or fabricate persona details
┌───────────────────────────────────────────────────────┐
│ Phase 1 - Planning (必须执行一次且只能一次) │
└───────────────────────────────────────────────────────┘
When persona READY and user gave a clear trend request:
1. Call write_todos EXACTLY ONCE with a strict plan containing:
- 3–6 concrete steps (numbered)
- Which URLs/topics to research
- Expected output of each major tool
- Final deliverable: one Markdown report
2. After receiving todos, you MUST follow this exact sequence unless impossible
3. Do NOT call any other tool until write_todos is done
┌───────────────────────────────────────────────────────┐
│ Phase 2 - Research & Collection │
└───────────────────────────────────────────────────────┘
Follow todos order:
- Use topic_research → get 3–8 high-quality URLs (add persona [Style] [Type] in query)
- Select best 3–6 URLs → call crawl4ai_batch ONCE with list
- Get file paths → call structured_retrieval ONCE with file_paths list
┌───────────────────────────────────────────────────────┐
│ Phase 3 - Synthesis & Delivery │
└───────────────────────────────────────────────────────┘
After structured_retrieval summary received:
- If extracted item count ≥ 8–12 AND covers main aspects in todos → ready to report
- Call report_generator ONCE (it reads local JSON/DB)
- After report_generator success → call terminate
- If data obviously insufficient → call topic_research again (max 1 extra round)
┌───────────────────────────────────────────────────────┐
│ HARD RULES - MUST OBEY │
└───────────────────────────────────────────────────────┘
• Never load full JSON/markdown into context - trust local storage
• Batch everything possible (crawl4ai_batch + structured_retrieval)
• Call tools in PHASE ORDER - no jumping, no repetition
• After report_generator → next action MUST be terminate
• If stuck > 4 steps without progress → call terminate with note "Incomplete - insufficient data"
• Never hallucinate trend data - base everything on retrieved content
• Report must start each section with **Conclusion First** insight
• Include [IMAGE_REF_xx] placeholders where visuals were extracted
Current status: Phase 0
"""

89
src/server/agent/run_test.py
View File

@@ -1,49 +1,74 @@
from langchain_core.messages import HumanMessage, AIMessage
import asyncio
from langchain_core.messages import HumanMessage, AIMessage, ToolMessage
from src.server.agent.graph import app
def main():
# 模拟 thread_id 区分不同用户或项目
async def async_main():
config = {"configurable": {"thread_id": "project_alpha"}}
print("測試模式已啟動 (輸入 'exit' 離開，'history' 查看歷史並回溯)")
use_report = input("是否启用深度报告？(y/n): ").lower() == 'y'
while True:
user_input = input("\n👤 设计师 (输入 'history' 定位轮次): ")
user_input = input("\n👤 輸入訊息: ").strip()
if user_input.lower() in ["exit", "quit", "結束"]:
print("測試結束")
break
# --- 官方推荐的异步回溯逻辑 ---
if user_input.lower() == "history":
print("\n--- 历史记录 ---")
for state in app.get_state_history(config):
# 每一个 state 都是一个 CheckpointTuple
cp_id = state.config["configurable"]["checkpoint_id"]
msg = state.values["messages"][-1].content[:30] if state.values.get("messages") else "Initial"
print(f"ID: {cp_id} | 内容: {msg}...")
target_id = input("\n请输入想要回溯的 Checkpoint ID (直接回车取消): ")
if target_id:
# 重新配置 config，指向特定的 checkpoint_id 实现分支
config = {"configurable": {"thread_id": "project_alpha", "checkpoint_id": target_id}}
print(f"✅ 已定位到节点 {target_id}，后续对话将从此分叉。")
# 你的 history 邏輯（這裡不變）
print("\n=== 歷史檢查點 ===")
states = [s async for s in app.aget_state_history(config)]
for idx, state_tuple in enumerate(states):
cp_id = state_tuple.config["configurable"].get("checkpoint_id", "N/A")
messages = state_tuple.values.get("messages", [])
if messages:
last_msg = messages[-1]
msg_type = type(last_msg).__name__
content_preview = str(last_msg.content)[:60].replace("\n", " ")
node = getattr(last_msg, "name", "無節點名")
print(f"[{idx}] {cp_id[:12]}... | {node} | {msg_type} | {content_preview}...")
target = input("\n輸入要回溯的 checkpoint ID (或 Enter 取消): ").strip()
if target:
config["configurable"]["checkpoint_id"] = target
print(f"已切換到 checkpoint {target}")
continue
# --- 官方推荐的 astream 异步流式调用 ---
print("🤖 Agent 思考中...")
for event in app.stream(
if not user_input:
continue
print("\n🤖 開始處理...")
try:
last_output = ""
async for event in app.astream(
{"messages": [HumanMessage(content=user_input)]},
config,
stream_mode="values" # 这里设为 values 可以直接获取当前状态的消息列表
stream_mode="updates"
):
# 获取当前节点处理后的最新消息
if "messages" in event:
last_msg = event["messages"][-1]
if isinstance(last_msg, AIMessage):
# 为了极致流式体验，可以在此处对 content 进行打印
pass
for node_name, update in event.items():
if "messages" in update:
for msg in update["messages"]:
if isinstance(msg, AIMessage):
content = msg.content.strip()
if content and content != last_output:
print(f"\n[{node_name}] {msg.name or 'AI'}: {content}")
last_output = content
elif isinstance(msg, ToolMessage):
print(f" → 工具 {msg.name}: {msg.content[:120]}{'...' if len(msg.content) > 120 else ''}")
else:
print(f" ({node_name}) {type(msg).__name__}")
# 运行结束后，最新的状态已经自动持久化到 MongoDB
# 我们可以通过 app.get_state(config) 验证
final_state = app.get_state(config)
print(f"\n✅ 最终回复: {final_state.values['messages'][-1].content}")
final_state = await app.aget_state(config)
final_msg = final_state.values["messages"][-1]
print(f"\n=== 完成 ===\n最終訊息: {final_msg.content[:300]}{'...' if len(final_msg.content) > 300 else ''}")
except Exception as e:
print(f"錯誤：{str(e)}")
import traceback
traceback.print_exc()
if __name__ == "__main__":
main()
asyncio.run(async_main())

3
src/server/agent/state.py
View File

@@ -1,7 +1,8 @@
import operator
from typing import Annotated, Sequence, TypedDict, Union
from typing import Annotated, Sequence, TypedDict, Union, Optional
from langchain_core.messages import BaseMessage
class AgentState(TypedDict):
# messages 存储完整的对话历史，operator.add 表示新消息是追加而不是覆盖
messages: Annotated[Sequence[BaseMessage], operator.add]

121
src/server/agent/tools/crawl_tool.py Normal file
View File

@@ -0,0 +1,121 @@
import time
import asyncio
from typing import List
from urllib.parse import urlparse
from pathlib import Path
from langchain_core.tools import tool
# ─────────────── 重要：計算路徑 ───────────────
# 目前這個檔案 (crawl4ai_batch.py) 所在的目錄
TOOL_DIR = Path(__file__).resolve().parent
# 專案根目錄（假設 tools 資料夾與主程式同級）
PROJECT_ROOT = TOOL_DIR.parent
# 儲存爬取結果的目錄（你可以自由決定放在哪裡）
# 建議選項 A：放在專案根目錄下的 workspace/raw_data
SAVE_DIR = PROJECT_ROOT / "workspace" / "raw_data"
# 建議選項 B：如果你打算讓 deep agent 直接讀取，建議放在 agent_workspace 底下
# SAVE_DIR = PROJECT_ROOT / "agent_workspace" / "raw_data"
# 確保目錄存在
SAVE_DIR.mkdir(parents=True, exist_ok=True)
# ────────────────────────────────────────────────
@tool
async def crawl4ai_batch(urls: List[str]) -> str:
"""
高性能网页爬虫，支持并行处理多个 URL。
爬取后的 Markdown 内容将保存到本地 workspace/raw_data 目录中。
返回执行结果摘要和保存的文件路径列表。
"""
if not urls:
return "❌ 错误: 未提供任何 URL。"
# print(f"🕷️ 正在并行爬取 {len(urls)} 个 URL...")
# print(f"儲存目錄: {SAVE_DIR}")
# Crawl4AI 配置（保持原樣）
from crawl4ai import AsyncWebCrawler, BrowserConfig, CrawlerRunConfig, CacheMode
browser_config = BrowserConfig(
headless=True,
verbose=False,
java_script_enabled=True,
user_agent="Mozilla/5.0 (Windows NT 10.0; Win64; x64) "
"AppleWebKit/537.36 (KHTML, like Gecko) "
"Chrome/118.0.5993.118 Safari/537.36",
proxy=None, # 可选，如果需要代理填 "http://user:pass@ip:port"
)
run_config = CrawlerRunConfig(
cache_mode=CacheMode.BYPASS,
word_count_threshold=5,
excluded_tags=["script", "style", "nav", "footer"],
remove_overlay_elements=True,
process_iframes=True,
)
results_summary = []
saved_files = []
try:
async with AsyncWebCrawler(config=browser_config) as crawler:
tasks = [crawler.arun(url=url, config=run_config) for url in urls]
crawl_results = await asyncio.gather(*tasks, return_exceptions=True)
for i, result in enumerate(crawl_results):
url = urls[i]
if isinstance(result, Exception):
results_summary.append(f"❌ 抓取失败 {url}: {str(result)}")
continue
if result.success:
markdown_content = result.markdown or ""
if len(markdown_content) < 500:
results_summary.append(f"⏩ 跳过 {url} (内容过短)")
continue
# 生成檔名
parsed = urlparse(url)
domain = parsed.netloc.replace("www.", "").replace(".", "_")
path_part = parsed.path.strip("/").replace("/", "_")[:50] or "index"
filename = f"{int(time.time())}_{domain}_{path_part}.md"
# 完整檔案路徑
filepath = SAVE_DIR / filename
# 寫入檔案
with open(filepath, "w", encoding="utf-8") as f:
header = f"<!-- Source: {url} -->\n<!-- Saved: {time.strftime('%Y-%m-%d %H:%M:%S')} -->\n\n"
f.write(header + markdown_content)
saved_files.append(str(filepath)) # 建議轉成字串
results_summary.append(f"✅ 成功: {url} → {filepath}")
else:
status = getattr(result, 'status_code', '未知错误')
results_summary.append(f"❌ 失败: {url} (状态码: {status})")
except Exception as e:
return f"🚨 爬虫系统崩溃: {str(e)}"
# 回傳給 agent 的結果
final_output = (
f"### 批量抓取完成 ###\n"
f"已成功保存 {len(saved_files)} 个文件。\n"
f"儲存目錄: {SAVE_DIR}\n"
f"详情:\n" + "\n".join(results_summary)
)
if saved_files:
final_output += "\n\n已保存的文件列表（可供後續讀取）:\n" + "\n".join(saved_files)
return final_output

47
src/server/agent/tools.py → src/server/agent/tools/generate_furniture_sketch.py
View File

@@ -1,11 +1,8 @@
import base64
import uuid
from google.oauth2 import service_account
from langchain_core.tools import tool
from google import genai
from google.genai.types import GenerateContentConfig, Modality
from PIL import Image
from io import BytesIO
from minio import Minio
@@ -27,21 +24,8 @@ client = genai.Client(
)
# --- 模拟你已经开发好的报告生成功能 ---
@tool
def generate_2025_report_tool(topic: str) -> str:
"""
专门用于收集信息并生成报告。
当用户询问关于趋势、市场分析、年度报告（如2025家具报告）时调用此工具。
"""
print(f"\n[系统日志] 正在调用外部模块生成关于 '{topic}' 的报告...")
# 这里对接你实际的代码，比如：return my_existing_module.run(topic)
return f"【报告生成成功】已生成关于 {topic} 的 PDF 报告。核心洞察：2025年趋势倾向于生物嗜好设计(Biophilic Design)和可持续软木材质。"
# --- 2. 绘图工具 (接入 Nano Banana 逻辑) ---
@tool
def generate_furniture_sketch(prompt: str) -> str:
def generate_furniture(prompt: str) -> str:
"""
使用 Gemini 图像生成模型根据详细的英文提示词生成家具设计草图。
"""
@@ -83,32 +67,3 @@ def generate_furniture_sketch(prompt: str) -> str:
return "图片生成成功，但上传至存储服务器失败。"
except Exception as e:
return f"绘图流程异常: {str(e)}"
if __name__ == '__main__':
print(generate_furniture_sketch("椅子"))
# creds = service_account.Credentials.from_service_account_file(
# settings.GOOGLE_GENAI_USE_VERTEXAI,
# scopes=["https://www.googleapis.com/auth/cloud-platform"],
# )
# client = genai.Client(
# credentials=creds,
# project=settings.GOOGLE_CLOUD_PROJECT,
# location=settings.GOOGLE_CLOUD_LOCATION,
# vertexai=True
# )
#
# response = client.models.generate_content(
# model="gemini-2.5-flash-image",
# contents=("Generate an image of the Eiffel tower with fireworks in the background."),
# config=GenerateContentConfig(
# response_modalities=[Modality.TEXT, Modality.IMAGE],
# ),
# )
#
# for part in response.candidates[0].content.parts:
# if part.text:
# print(part.text)
# elif part.inline_data:
# image = Image.open(BytesIO((part.inline_data.data)))
# image.save("example-image-eiffel-tower.png")

36
src/server/agent/tools/read_file.py Normal file
View File

@@ -0,0 +1,36 @@
import os
from langchain_core.tools import tool
@tool
def read_file(file_path: str) -> str:
"""
读取本地文件的万能工具。支持绝对路径和相对路径。
"""
# 1. 极端清洗：去掉 Agent 可能误加的引号、空格或转义符
path = file_path.strip().strip("'").strip('"').replace("\\", "/")
# 2. 打印当前环境真相（在你的 Python 控制台可见）
print(f"\n--- 🛠️ READ_FILE 调试信息 ---")
print(f"待读路径: {path}")
print(f"当前工作目录 (CWD): {os.getcwd()}")
print(f"是否存在: {os.path.exists(path)}")
# 3. 尝试直接读取（跳过任何沙箱逻辑）
try:
# 如果是相对路径，尝试转为绝对路径再读
abs_path = os.path.abspath(path)
if os.path.exists(abs_path):
with open(abs_path, 'r', encoding='utf-8') as f:
content = f.read()
return content
else:
# 如果读不到，列出父目录内容作为线索
parent = os.path.dirname(abs_path)
if os.path.exists(parent):
files = os.listdir(parent)
return f"错误：文件不存在。该目录下现有的文件有: {files[:5]}..."
return f"错误：路径不存在，且连父目录 {parent} 都找不到。"
except Exception as e:
return f"读取失败，系统异常: {str(e)}"

157
src/server/agent/tools/report_generator_tool.py Normal file
View File

@@ -0,0 +1,157 @@
import os
import json
import re
from typing import Optional, List, Dict
from langchain_qwq import ChatQwen
from pydantic import BaseModel, Field
from langchain_core.tools import tool
from langchain_core.messages import SystemMessage, HumanMessage
from src.core.config import settings
# =========================
# LLM 初始化
# =========================
llm = ChatQwen(
model="qwen3.5-flash",
temperature=0.2,
max_tokens=3_000,
timeout=None,
max_retries=2,
api_key=settings.QWEN_API_KEY)
# =========================
# Tool 输入 Schema
# =========================
class ReportInput(BaseModel):
report_topic: str = Field(
...,
description="Main topic of the report, e.g. '2026 Sofa Design Trends'"
)
structured_data: List[Dict] = Field(
...,
description="Structured retrieval result items"
)
language: Optional[str] = Field(
default="English",
description="Output language"
)
# =========================
# LangGraph Tool
# =========================
@tool("report_generator", args_schema=ReportInput)
async def report_generator(
report_topic: str,
structured_data: List[Dict],
language: str = "English"
) -> dict:
"""
Generate a professional design/market report
directly from structured retrieval results.
"""
if not structured_data:
return {
"status": "error",
"message": "No structured data provided."
}
collected_data_str = json.dumps(
structured_data,
ensure_ascii=False,
indent=2
)
# =========================
# Prompt
# =========================
system_prompt = f"""
You are a professional design trend analyst.
Generate a long, structured Markdown report.
REQUIREMENTS:
1. Follow MECE principle.
2. Embed images ONLY if they start with https://
using: ![alt](url)
3. Insert images inline.
4. Every key insight must cite source:
[Website Name](url)
5. Use Markdown headings.
6. Start directly with title.
7. Be detailed and analytical.
Output Language: {language}
"""
user_prompt = f"""
Topic: {report_topic}
Input Data:
{collected_data_str}
"""
# =========================
# 调用 LLM
# =========================
try:
response = await llm.ainvoke([
SystemMessage(content=system_prompt),
HumanMessage(content=user_prompt)
])
report_content = response.content.strip()
# 清理 markdown block 包裹
report_content = (
report_content
.replace("```markdown", "")
.replace("```", "")
.strip()
)
except Exception as e:
return {
"status": "error",
"message": f"LLM generation failed: {str(e)}"
}
# =========================
# 保存报告
# =========================
output_dir = "workspace/reports"
os.makedirs(output_dir, exist_ok=True)
safe_topic = re.sub(
r'[\\/*?:"<>|]',
"",
report_topic.replace(" ", "_")
)
filename = f"{output_dir}/{safe_topic}.md"
try:
with open(filename, "w", encoding="utf-8") as f:
f.write(report_content)
except Exception as e:
return {
"status": "error",
"message": f"Failed to save report: {str(e)}"
}
return {
"status": "success",
"file_path": filename,
"message": "Report generated successfully."
}

74
src/server/agent/tools/research_tool.py Normal file
View File

@@ -0,0 +1,74 @@
import asyncio
import json
from datetime import datetime
from typing import List, Set, Optional
from langchain_core.tools import tool
from tavily import TavilyClient
from src.core.config import settings
# 模拟配置加载
TAVILY_API_KEY = settings.TAVILY_API_KEY
@tool
async def topic_research(topic: str, max_urls: int = 15) -> str:
"""
深度调研工具。该工具会利用 Tavily 搜索引擎针对特定主题进行多维度搜索。
它会自动生成针对性的搜索词（包含年份和趋势），并返回去重后的高质量 URL 列表。
"""
if not TAVILY_API_KEY:
return "❌ 错误: 未配置 TAVILY_API_KEY。"
client = TavilyClient(api_key=TAVILY_API_KEY)
# 1. 自动生成多维度搜索词 (在工具内部快速生成)
current_year = datetime.now().strftime("%Y")
queries = [
f"{topic} trends {current_year}",
f"{topic} market analysis {current_year}",
f"top selling {topic} styles {current_year}",
f"best {topic} materials and colors {current_year}"
]
# 2. 并行执行搜索
async def perform_search(q: str):
# 使用 asyncio.to_thread 运行同步的 Tavily SDK
def sync_search():
try:
response = client.search(
query=q,
search_depth="advanced",
max_results=5,
include_answer=False
)
return response.get('results', [])
except Exception as e:
print(f"Search error: {e}")
return []
return await asyncio.to_thread(sync_search)
search_tasks = [perform_search(q) for q in queries]
search_results_list = await asyncio.gather(*search_tasks)
# 3. 结果去重与过滤
seen_urls: Set[str] = set()
final_urls = []
# 常见的非内容页面过滤
skip_extensions = ('.pdf', '.jpg', '.png', '.zip', '.exe')
for results in search_results_list:
for item in results:
url = item.get('url')
if url and url not in seen_urls:
if not url.lower().endswith(skip_extensions):
seen_urls.add(url)
final_urls.append(url)
# 4. 结果截断
selected_urls = final_urls[:max_urls]
# 返回 JSON 字符串，便于 Agent 下一步调用批量爬虫 (Crawl4ai)
return json.dumps(selected_urls, ensure_ascii=False)

27
src/server/agent/tools/save_to_local.py Normal file
View File

@@ -0,0 +1,27 @@
import os
from langchain_core.tools import tool
# 定义本地保存路径
OUTPUT_DIR = "./research_reports"
if not os.path.exists(OUTPUT_DIR):
os.makedirs(OUTPUT_DIR)
@tool
def save_to_local_disk(filename: str, content: str) -> str:
"""
将内容保存到本地物理磁盘。
filename: 文件名（例如 'sofa_report.md'）
content: 调研报告或数据的文本内容
"""
try:
# 移除非法路径字符，确保安全
safe_filename = os.path.basename(filename)
file_path = os.path.join(OUTPUT_DIR, safe_filename)
with open(file_path, "w", encoding="utf-8") as f:
f.write(content)
return f"✅ 成功！文件已保存至本地物理路径: {os.path.abspath(file_path)}"
except Exception as e:
return f"❌ 保存失败，错误原因: {str(e)}"

225
src/server/agent/tools/structured_retrieval_tool.py Normal file
View File

@@ -0,0 +1,225 @@
import os
import re
import json
from datetime import datetime
from typing import List, Dict, Optional
from pydantic import BaseModel, Field
from langchain_core.tools import tool
from langchain_core.documents import Document
# RAG
from langchain_community.vectorstores import FAISS
from langchain_huggingface import HuggingFaceEmbeddings
from sentence_transformers import CrossEncoder
# =========================
# 全局模型（单例）
# =========================
_EMBEDDING_MODEL = HuggingFaceEmbeddings(
model_name="sentence-transformers/all-MiniLM-L6-v2"
)
_RERANK_MODEL = CrossEncoder(
"cross-encoder/ms-marco-MiniLM-L-6-v2"
)
class StructuredRetrievalInput(BaseModel):
file_paths: List[str] = Field(..., description="List of local markdown file paths.")
query: str = Field(..., description="Extraction query")
source_url: Optional[str] = Field(None, description="Optional global source URL")
@tool("structured_retrieval", args_schema=StructuredRetrievalInput)
def structured_retrieval(
file_paths: List[str],
query: str,
source_url: Optional[str] = None
) -> Dict:
"""
Batch structured extraction from markdown files.
- Performs vector search + re-ranking
- Saves extracted structured data as JSON file to disk
- Returns ONLY summary (status, count, file path)
"""
# ── 1. 收集所有文件內容 ──────────────────────────────────────
all_docs_pool: List[Document] = []
for path in file_paths:
if not os.path.exists(path) or not path.endswith((".md", ".markdown")):
continue
file_name = os.path.basename(path)
with open(path, "r", encoding="utf-8") as f:
content = f.read()
current_source = source_url or _extract_source_from_md(content) or "unknown"
sections = _split_markdown_by_headers(content)
for sec in sections:
all_docs_pool.append(
Document(
page_content=sec,
metadata={"source_url": current_source, "file_name": file_name}
)
)
if not all_docs_pool:
return {"status": "no_documents_found", "items_count": 0, "json_path": None}
# ── 2. Vector search ────────────────────────────────────────────
vector_store = FAISS.from_documents(all_docs_pool, _EMBEDDING_MODEL)
retrieved = vector_store.similarity_search(query, k=200)
# ── 3. 提取結構化片段 ──────────────────────────────────────────
structured_items = []
for doc in retrieved:
text = doc.page_content.strip()
if len(text) < 30:
continue
images = list(set(re.findall(r"!\[.*?\]\((.*?)\)", text)))
structured_items.append(
{
"text": text,
"images": images,
"source_url": doc.metadata.get("source_url"),
"file_name": doc.metadata.get("file_name")
}
)
# ── 4. Re-rank ──────────────────────────────────────────────────
if structured_items:
unique_items = {item["text"]: item for item in structured_items}.values()
pairs = [[query, item["text"]] for item in unique_items]
scores = _RERANK_MODEL.predict(pairs)
sorted_items = sorted(
zip(scores, unique_items),
key=lambda x: x[0],
reverse=True
)
top_items = [item for _, item in sorted_items[:50]]
else:
top_items = []
# ── 5. 寫入 JSON 文件 ──────────────────────────────────────────
if not top_items:
return {"status": "no_relevant_content", "items_count": 0, "json_path": None}
# 產生有意義的檔名
safe_query = re.sub(r'[^a-zA-Z0-9\u4e00-\u9fa5]', '_', query)[:40]
timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
json_filename = f"extracted_{safe_query}_{timestamp}.json"
# 建議的儲存目錄（與 crawl4ai_batch 對齊）
output_dir = os.path.join(os.path.dirname(file_paths[0]), "..", "extracted")
os.makedirs(output_dir, exist_ok=True)
json_path = os.path.join(output_dir, json_filename)
with open(json_path, "w", encoding="utf-8") as f:
json.dump(
{
"query": query,
"extracted_at": timestamp,
"item_count": len(top_items),
"items": top_items
},
f,
ensure_ascii=False,
indent=2
)
# ── 6. 只回傳摘要 ──────────────────────────────────────────────
return {
"status": "success",
"items_count": len(top_items),
"json_path": json_path,
"summary": f"已提取 {len(top_items)} 個高相關片段，儲存於 {json_path}"
}
def _extract_source_from_md(content: str) -> Optional[str]:
match = re.search(r"<!--\s*Source:\s*(.*?)\s*-->", content)
return match.group(1).strip() if match else None
# =========================
# Markdown Header Split
# =========================
def _split_markdown_by_headers(
content: str,
max_chars: int = 2000,
overlap: int = 150,
):
header_re = re.compile(
r'^(#{1,6})\s+(.+?)\s*$',
re.MULTILINE
)
matches = list(header_re.finditer(content))
if not matches:
return _chunk_text(content, max_chars, overlap)
sections = []
for i, m in enumerate(matches):
start = m.start()
end = (
matches[i + 1].start()
if i + 1 < len(matches)
else len(content)
)
block = content[start:end].strip()
if block:
sections.append(block)
final_sections = []
for s in sections:
if len(s) > max_chars:
final_sections.extend(
_chunk_text(s, max_chars, overlap)
)
else:
final_sections.append(s)
return final_sections
def _chunk_text(
text: str,
max_chars: int = 2000,
overlap: int = 150
):
text = text.strip()
if len(text) <= max_chars:
return [text]
chunks = []
start = 0
while start < len(text):
end = min(len(text), start + max_chars)
chunk = text[start:end].strip()
if chunk:
chunks.append(chunk)
if end == len(text):
break
start = max(0, end - overlap)
return chunks

38
src/server/agent/tools/terminate_tool.py Normal file
View File

@@ -0,0 +1,38 @@
from typing import Literal
from langchain_core.tools import tool
from pydantic import BaseModel, Field
class TerminateInput(BaseModel):
"""終止對話的輸入參數"""
status: Literal["success", "failure"] = Field(
description="互動結束的狀態：'success' 表示任務完成，'failure' 表示無法繼續",
examples=["success", "failure"]
)
reason: str = Field(
default="",
description="可選：簡單說明為什麼結束（例如 '報告已生成' 或 '缺少關鍵資訊'）",
examples=["報告已成功生成", "無法取得足夠資料"]
)
@tool(args_schema=TerminateInput)
def terminate(status: str, reason: str = "") -> str:
"""
當任務完成、報告已生成，或無法繼續進行時，呼叫此工具來結束本次互動。
使用時機：
- 已經成功產生最終報告（report_generator 已完成）
- 遇到無法解決的錯誤或缺少關鍵資訊
- 用戶需求已完全滿足
請在呼叫前確保所有必要步驟已完成，並在 reason 中簡單說明結束原因。
"""
if status not in ("success", "failure"):
status = "failure" # 防呆
msg = f"互動已終止，狀態：{status.upper()}"
if reason:
msg += f"\n原因：{reason}"
return msg

96
src/server/agent/tools/user_persona_tool.py Normal file
View File

@@ -0,0 +1,96 @@
import json
import os
from typing import List, Literal, Optional, Dict, Any
from langchain_core.tools import tool
# 定义存储路径
DB_PATH = os.path.join("workspace", "user_persona.json")
def _load_store() -> Dict[str, Any]:
"""从本地文件加载画像数据"""
if os.path.exists(DB_PATH):
try:
with open(DB_PATH, "r", encoding="utf-8") as f:
return json.load(f)
except Exception:
return {}
return {}
def _save_store(data: Dict[str, Any]):
"""将画像数据保存到本地文件"""
os.makedirs(os.path.dirname(DB_PATH), exist_ok=True)
with open(DB_PATH, "w", encoding="utf-8") as f:
json.dump(data, f, ensure_ascii=False, indent=2)
@tool
def manage_user_persona(
command: Literal["set", "update", "get", "clear"],
design_type: Optional[str] = None,
style_preference: Optional[str] = None,
budget_range: Optional[str] = None,
color_palette: Optional[List[str]] = None,
target_audience: Optional[str] = None,
extra_requirements: Optional[str] = None
) -> str:
"""
用户画像与设计偏好管理工具。
用于设定、更新、获取或重置用户的设计上下文（如风格、预算、颜色）。
Agent 在开始调研前必须先调用 get 获取画像，若关键信息缺失需引导用户补充。
"""
# 每次调用都重新读取，确保多进程或重启后数据一致
store = _load_store()
if command == "clear":
if os.path.exists(DB_PATH):
os.remove(DB_PATH)
return "✅ 用户个性化模板已从本地文件清空。"
if command == "get":
if not store:
return "⚠️ [缺失信息] 当前尚未配置画像。请询问用户：设计类型（如沙发）、风格偏好（如极简）等。"
# 格式化输出供 Agent 阅读
res = [
"--- 👤 实时用户画像 (本地存储) ---",
f"🎯 类型: {store.get('design_type', '未设定')}",
f"🎨 风格: {store.get('style_preference', '未设定')}",
f"💰 预算: {store.get('budget_range', '未设定')}",
f"🌈 色系: {', '.join(store.get('color_palette', [])) or '未设定'}",
f"👥 受众: {store.get('target_audience', '未设定')}",
f"📝 需求: {store.get('extra_requirements', '未设定')}",
"-----------------------"
]
# 逻辑检查
if not store.get('design_type') or not store.get('style_preference'):
res.append("\n⚠️ 关键信息缺失，建议补充 '设计类型' 和 '风格偏好'。")
return "\n".join(res)
if command in ["set", "update"]:
if command == "set":
store = {} # 重置内存中的字典
# 提取传入的非空参数
update_data = {
"design_type": design_type,
"style_preference": style_preference,
"budget_range": budget_range,
"color_palette": color_palette,
"target_audience": target_audience,
"extra_requirements": extra_requirements
}
# 更新有效字段
for k, v in update_data.items():
if v is not None:
store[k] = v
# 保存到文件
_save_store(store)
return f"✅ 本地画像已同步。当前配置：\n{json.dumps(store, ensure_ascii=False, indent=2)}"
return "❌ 错误：未知命令。"

3010
uv.lock generated
View File

File diff suppressed because it is too large Load Diff