修改知识库范围

.env.template

@@ -2,22 +2,22 @@
 # DB_URL=postgresql+psycopg://ai:ai@localhost:5532/ai
 
 # OpenAI API配置
-API_KEY=sk-oftybjqntjpxlhkcxkieluljb
+API_KEY=sk-oftybjqntjpxlhkc
-EMBEDDING_MODEL=BAAI/bge-large-zh-v1.5
+EMBEDDING_MODEL=bge-m3
-EMBEDDING_BASE_URL=https://api.siliconflow.cn/v1
+EMBEDDING_BASE_URL=http://10.1.16.39:9995/v1
-MODEL_LIST=Qwen2.5-72B=openai:Qwen/Qwen2.5-72B-Instruct&gpt-4o=openai:gpt-4o
+MODEL_LIST=Qwen2.5-72B=openai:Qwen2.5-72B-Instruct-GPTQ-Int8
-
+MODEL_BASE_URL=http://172.20.0.145:9995/v1
-# 模型配置
-MODEL_BASE_URL=https://api.siliconflow.cn/v1
 
 # 文件路径配置
-KNOWLEDGE_SOURCE_DIR=data
+MINGCI_KNOWLEDGE_SOURCE_DIR=data/业务名词库
-MEMORY_DB_FILE=tmp/agent_memory.db
+MINGCI_VECTOR_DB_PATH=tmp/mingcidb
-VECTOR_DB_PATH=tmp/lancedb
-SESSION_STORAGE_PATH=tmp/agent_sessions_json
 
-# 知识库加载控制
+KNOWLEDGE_SOURCE_DIR=data/控件布局
-LOAD_KNOWLEDGE=true
+VECTOR_DB_PATH=tmp/knowledgedb
+
+MEMORY_DB_FILE=tmp/agent_memory.db
+
+SESSION_STORAGE_PATH=tmp/agent_sessions_json
 
 AGNO_MONITOR=true
 AGNO_TELEMETRY=true

agentic_rag.py

@@ -31,6 +31,7 @@
 from pathlib import Path
 
 from agno.document.chunking.document import DocumentChunking
+from agno.models.deepseek import DeepSeek
 from dotenv import load_dotenv
 
 # 加载.env文件
@@ -107,6 +108,8 @@ def get_model_by_provider(provider: str, model_name: str):
     #     return Claude(id=model_name)
     # elif provider == "groq":
     #     return Groq(id=model_name)
+    elif provider == "deepseek":
+         return DeepSeek(id=model_name, base_url=model_baseUrl, api_key=api_key)
     else:
         raise ValueError(f"Unsupported model provider: {provider}")
         
@@ -127,8 +130,17 @@ def initialize_memory(model) -> AgentMemory:
 def initialize_vector_db() -> LanceDb:
     """初始化并返回配置好的LanceDb实例"""
     return LanceDb(
-        table_name="recipes",
+        table_name="knowledge",
-        uri=os.getenv("VECTOR_DB_PATH", "tmp/lancedb"),
+        uri=os.getenv("VECTOR_DB_PATH", "tmp/knowledgedb"),
+        search_type=SearchType.hybrid,
+        embedder=OpenAIEmbedder(id=embedding_model, base_url=embedding_baseUrl, api_key=api_key)
+    )
+
+def initialize_mingci_vector_db() -> LanceDb:
+    """初始化并返回配置好的LanceDb实例"""
+    return LanceDb(
+        table_name="mingci",
+        uri=os.getenv("MINGCI_VECTOR_DB_PATH", "tmp/mingcidb"),
         search_type=SearchType.hybrid,
         embedder=OpenAIEmbedder(id=embedding_model, base_url=embedding_baseUrl, api_key=api_key)
     )
@@ -146,6 +158,18 @@ def initialize_knowledge_base() -> AgentKnowledge:
         reader=TextReader(),  # 默认文本读取器
     )
 
+def initialize_mingci_knowledge_base() -> AgentKnowledge:
+    """初始化并返回配置好的AgentKnowledge实例"""
+    return AgentKnowledge(
+        vector_db=initialize_mingci_vector_db(),
+        num_documents=3,  # 检索3个最相关的文档
+        chunking_strategy=DocumentChunking(
+            chunk_size=500,
+            overlap=50,
+        ),  # 固定大小分块
+        optimize_on=1000,  # 每1000条数据进行向量优化
+        reader=TextReader(),  # 默认文本读取器
+    )
 
 def get_agentic_rag_agent(
     model_id: str = "openai:gpt-4o",

main.py

@@ -5,7 +5,7 @@ from agno.document import Document
 from agno.utils.log import logger
 from dotenv import load_dotenv
 
-from agentic_rag import initialize_knowledge_base, get_reader
+from agentic_rag import initialize_knowledge_base, get_reader, initialize_mingci_knowledge_base
 
 # 加载.env文件
 load_dotenv()
@@ -14,25 +14,34 @@ import os
 def main():
     print("Hello from agno-agentic-rag!")
     # 从.env加载知识库来源目录并初始化知识库
-    load_knowledge = os.getenv("LOAD_KNOWLEDGE", "false").lower() == "true"
+    mingci_knowledge_source_dir = os.getenv("MINGCI_KNOWLEDGE_SOURCE_DIR")
+    if mingci_knowledge_source_dir and os.path.exists(mingci_knowledge_source_dir):
+        # 初始化知识库
+        knowledge_base = initialize_mingci_knowledge_base()
+
+        LoadKnowledgeToDatabase(knowledge_base, mingci_knowledge_source_dir)
+
     knowledge_source_dir = os.getenv("KNOWLEDGE_SOURCE_DIR")
-    if load_knowledge and knowledge_source_dir and os.path.exists(knowledge_source_dir):
+    if knowledge_source_dir and os.path.exists(knowledge_source_dir):
         # 初始化知识库
         knowledge_base = initialize_knowledge_base()
 
-        logger.info(f"加载知识库: {knowledge_source_dir}")
+        LoadKnowledgeToDatabase(knowledge_base, knowledge_source_dir)
-        for root, _, files in os.walk(knowledge_source_dir):
+
-            for file in files:
+def LoadKnowledgeToDatabase(knowledge_base, knowledge_source_dir):
-                file_path = os.path.join(root, file)
+    logger.info(f"加载知识库: {knowledge_source_dir}")
-                file_ext = os.path.splitext(file)[1][1:]  # 获取文件扩展名
+    for root, _, files in os.walk(knowledge_source_dir):
-                reader = get_reader(file_ext)
+        for file in files:
-                if reader:
+            file_path = os.path.join(root, file)
-                    try:
+            file_ext = os.path.splitext(file)[1][1:]  # 获取文件扩展名
-                        filePath = Path(file_path)
+            reader = get_reader(file_ext)
-                        docs: List[Document] = reader.read(filePath)
+            if reader:
-                        knowledge_base.load_documents(docs, upsert=True)
+                try:
-                    except Exception as e:
+                    filePath = Path(file_path)
-                        logger.warning(f"无法加载文档 {file_path}: {str(e)}")
+                    docs: List[Document] = reader.read(filePath)
+                    knowledge_base.load_documents(docs, upsert=True)
+                except Exception as e:
+                    logger.warning(f"无法加载文档 {file_path}: {str(e)}")
 
 
 if __name__ == "__main__":

pyproject.toml

@@ -11,6 +11,7 @@ dependencies = [
     "nest-asyncio>=1.6.0",
     "streamlit>=1.44.1",
     "openai",
+    "pylance",
     "extra-streamlit-components>=0.1.71",
     "sqlalchemy>=2.0.38",
     "websockets>=14.2",