From 21a5fdf930e280610b16e8213921e8970cb0b97a Mon Sep 17 00:00:00 2001
From: panda <1415243231@qq.com>
Date: Thu, 14 May 2026 23:20:56 +0800
Subject: [PATCH] =?UTF-8?q?feat:=20=E5=90=8E=E7=AB=AF=E5=9F=BA=E7=A1=80?=
 =?UTF-8?q?=E8=AE=BE=E6=96=BD=20=E2=80=94=20LLM=E5=B7=A5=E5=8E=82/Embeddin?=
 =?UTF-8?q?g=E5=B7=A5=E5=8E=82/=E9=AA=8C=E8=AF=81=E5=AE=A2=E6=88=B7?=
 =?UTF-8?q?=E7=AB=AF/=E4=BC=9A=E8=AF=9D=E6=8C=81=E4=B9=85=E5=8C=96?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

- backend/llm.py: 支持 OpenAI 兼容 API 与 Ollama 本地模型切换
- backend/embeddings.py: 支持云端与本地嵌入模型（sentence-transformers）
- backend/validation.py: FastAPI 验证服务 HTTP 客户端
- backend/session.py: JSON 文件会话管理（创建/加载/保存/列表/删除）
- .env.example: 完整环境变量模板
- requirements.txt: 所有 Python 依赖声明
---
 .env.example          |  38 ++++++++++++++
 .gitignore            |  23 +++++++++
 backend/__init__.py   |   0
 backend/embeddings.py |  26 ++++++++++
 backend/llm.py        |  28 +++++++++++
 backend/session.py    | 112 ++++++++++++++++++++++++++++++++++++++++++
 backend/validation.py |  26 ++++++++++
 requirements.txt      |  27 ++++++++++
 8 files changed, 280 insertions(+)
 create mode 100644 .env.example
 create mode 100644 .gitignore
 create mode 100644 backend/__init__.py
 create mode 100644 backend/embeddings.py
 create mode 100644 backend/llm.py
 create mode 100644 backend/session.py
 create mode 100644 backend/validation.py
 create mode 100644 requirements.txt

diff --git a/.env.example b/.env.example
new file mode 100644
index 0000000..35668fb
--- /dev/null
+++ b/.env.example
@@ -0,0 +1,38 @@
+# 大语言模型后端：cloud 或 local
+LLM_BACKEND=cloud
+
+# 云端配置（OpenAI 兼容）
+OPENAI_API_KEY=sk-xxxx
+OPENAI_BASE_URL=https://api.openai.com/v1
+LLM_MODEL=gpt-4o
+
+# 本地大语言模型（Ollama）
+LOCAL_LLM_MODEL=qwen2.5-coder:7b
+
+# 嵌入模型后端：local 或 cloud
+EMBED_BACKEND=local
+LOCAL_EMBED_MODEL=Qwen/Qwen3-Embedding-0.6B
+
+# 验证服务地址
+VALIDATION_SERVICE_URL=http://localhost:8001/validate
+
+# Chroma 持久化目录
+CHROMA_PERSIST_DIR=./db/chroma
+
+# 最大自动修正尝试次数
+MAX_RETRY=3
+
+# 上下文压缩阈值（token 数）
+CONTEXT_MAX_TOKENS=6000
+
+# 保留最近 N 轮完整对话
+CONTEXT_KEEP_RECENT=4
+
+# 会话持久化目录
+SESSIONS_DIR=./sessions
+
+# 状态快照保留数量（用于撤销操作）
+HISTORY_MAX_SNAPSHOTS=10
+
+# 意图识别模型（默认使用主 LLM 模型）
+# INTENT_MODEL=gpt-4o-mini
diff --git a/.gitignore b/.gitignore
new file mode 100644
index 0000000..fda7982
--- /dev/null
+++ b/.gitignore
@@ -0,0 +1,23 @@
+# 环境配置（含密钥）
+.env
+
+# Python
+__pycache__/
+*.py[cod]
+*.egg-info/
+dist/
+.venv/
+
+# 数据库
+db/chroma/
+sessions/
+
+# IDE
+.idea/
+.vscode/
+*.swp
+*.swo
+
+# 系统文件
+Thumbs.db
+.DS_Store
diff --git a/backend/__init__.py b/backend/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/backend/embeddings.py b/backend/embeddings.py
new file mode 100644
index 0000000..0c7ee93
--- /dev/null
+++ b/backend/embeddings.py
@@ -0,0 +1,26 @@
+"""嵌入模型工厂：支持本地 sentence-transformers 和云端 API。"""
+
+import os
+from dotenv import load_dotenv
+
+load_dotenv()
+
+
+def get_embeddings():
+    backend = os.getenv("EMBED_BACKEND", "local")
+    if backend == "cloud":
+        from langchain_openai import OpenAIEmbeddings
+
+        return OpenAIEmbeddings(
+            model=os.getenv("EMBED_CLOUD_MODEL", "text-embedding-3-small"),
+            api_key=os.getenv("OPENAI_API_KEY"),
+            base_url=os.getenv("OPENAI_BASE_URL", "https://api.openai.com/v1"),
+        )
+    else:
+        try:
+            from langchain_huggingface import HuggingFaceEmbeddings
+        except ImportError:
+            from langchain_community.embeddings import HuggingFaceEmbeddings
+
+        model = os.getenv("LOCAL_EMBED_MODEL", "Qwen/Qwen3-Embedding-0.6B")
+        return HuggingFaceEmbeddings(model_name=model)
diff --git a/backend/llm.py b/backend/llm.py
new file mode 100644
index 0000000..a9584ec
--- /dev/null
+++ b/backend/llm.py
@@ -0,0 +1,28 @@
+"""大语言模型工厂：支持 OpenAI 兼容的云端 API 和本地 Ollama。"""
+
+import os
+from dotenv import load_dotenv
+
+load_dotenv()
+
+
+def get_llm():
+    backend = os.getenv("LLM_BACKEND", "cloud")
+    if backend == "local":
+        from langchain_ollama import ChatOllama
+
+        model = os.getenv("LOCAL_LLM_MODEL", "qwen2.5-coder:7b")
+        return ChatOllama(model=model, temperature=0.1)
+    else:
+        from langchain_openai import ChatOpenAI
+
+        return ChatOpenAI(
+            model=os.getenv("LLM_MODEL", "gpt-4o"),
+            api_key=os.getenv("OPENAI_API_KEY"),
+            base_url=os.getenv("OPENAI_BASE_URL", "https://api.openai.com/v1"),
+            temperature=0.1,
+        )
+
+
+def get_llm_for_correction():
+    return get_llm()
diff --git a/backend/session.py b/backend/session.py
new file mode 100644
index 0000000..8a41d58
--- /dev/null
+++ b/backend/session.py
@@ -0,0 +1,112 @@
+"""多会话持久化管理模块。
+
+每个会话对应一个独立的 JSON 文件存储在 ./sessions/ 目录下。
+"""
+
+import json
+import os
+import uuid
+from datetime import datetime, timezone
+from pathlib import Path
+from typing import Optional
+
+from dotenv import load_dotenv
+
+load_dotenv()
+
+SESSIONS_DIR = Path(os.getenv("SESSIONS_DIR", "./sessions"))
+
+
+def _ensure_dir():
+    SESSIONS_DIR.mkdir(parents=True, exist_ok=True)
+
+
+def _session_path(session_id: str) -> Path:
+    return SESSIONS_DIR / f"{session_id}.json"
+
+
+def generate_session_id() -> str:
+    return uuid.uuid4().hex[:12]
+
+
+def create_session(name: str = "", agent_state: Optional[dict] = None) -> dict:
+    """创建新会话，返回会话元数据。"""
+    _ensure_dir()
+    sid = generate_session_id()
+    now = _now_iso()
+    data = {
+        "session_id": sid,
+        "session_name": name or f"新建报表 {now[:10]}",
+        "created_at": now,
+        "updated_at": now,
+        "agent_state": agent_state or {},
+    }
+    with open(_session_path(sid), "w", encoding="utf-8") as f:
+        json.dump(data, f, ensure_ascii=False, indent=2)
+    return data
+
+
+def load_session(session_id: str) -> Optional[dict]:
+    """按 ID 加载会话数据。未找到则返回 None。"""
+    _ensure_dir()
+    fp = _session_path(session_id)
+    if not fp.exists():
+        return None
+    with open(fp, "r", encoding="utf-8") as f:
+        return json.load(f)
+
+
+def save_session(session_id: str, agent_state: dict, session_name: str = ""):
+    """将会话状态保存（更新）至磁盘。"""
+    _ensure_dir()
+    fp = _session_path(session_id)
+    data = {}
+    if fp.exists():
+        with open(fp, "r", encoding="utf-8") as f:
+            data = json.load(f)
+
+    data["session_id"] = session_id
+    if session_name:
+        data["session_name"] = session_name
+    if not data.get("session_name"):
+        data["session_name"] = f"报表 {data.get('created_at', _now_iso())[:10]}"
+    data["updated_at"] = _now_iso()
+    if not data.get("created_at"):
+        data["created_at"] = data["updated_at"]
+    data["agent_state"] = agent_state
+
+    with open(fp, "w", encoding="utf-8") as f:
+        json.dump(data, f, ensure_ascii=False, indent=2)
+
+
+def list_all_sessions() -> list[dict]:
+    """列出所有历史会话（仅摘要，不含完整 agent_state）。"""
+    _ensure_dir()
+    sessions = []
+    for fp in sorted(SESSIONS_DIR.glob("*.json"), key=os.path.getmtime, reverse=True):
+        try:
+            with open(fp, "r", encoding="utf-8") as f:
+                data = json.load(f)
+            sessions.append({
+                "session_id": data.get("session_id", fp.stem),
+                "session_name": data.get("session_name", fp.stem),
+                "created_at": data.get("created_at", ""),
+                "updated_at": data.get("updated_at", ""),
+            })
+        except (json.JSONDecodeError, KeyError):
+            continue
+    return sessions
+
+
+def delete_session(session_id: str) -> bool:
+    """按 ID 删除会话文件。"""
+    _ensure_dir()
+    fp = _session_path(session_id)
+    if fp.exists():
+        fp.unlink()
+        return True
+    return False
+
+
+def _now_iso() -> str:
+    return datetime.now(timezone.utc).isoformat()
diff --git a/backend/validation.py b/backend/validation.py
new file mode 100644
index 0000000..df7647d
--- /dev/null
+++ b/backend/validation.py
@@ -0,0 +1,26 @@
+"""FastAPI 验证服务的客户端。"""
+
+import os
+
+import httpx
+from dotenv import load_dotenv
+
+load_dotenv()
+
+VALIDATION_URL = os.getenv("VALIDATION_SERVICE_URL", "http://localhost:8001/validate")
+
+
+def validate_jrxml(jrxml_text: str) -> dict:
+    """将 JRXML 发送到验证服务并返回 {valid: bool, error: str}。"""
+    try:
+        with httpx.Client(timeout=30.0) as client:
+            resp = client.post(VALIDATION_URL, json={"jrxml": jrxml_text})
+            resp.raise_for_status()
+            return resp.json()
+    except httpx.ConnectError:
+        return {
+            "valid": False,
+            "error": f"无法连接到验证服务 ({VALIDATION_URL})。是否正在运行？",
+        }
+    except Exception as e:
+        return {"valid": False, "error": f"验证请求失败: {str(e)}"}
diff --git a/requirements.txt b/requirements.txt
new file mode 100644
index 0000000..adb3b65
--- /dev/null
+++ b/requirements.txt
@@ -0,0 +1,27 @@
+# 核心依赖
+streamlit>=1.28.0
+langgraph>=0.2.0
+langchain>=0.3.0
+langchain-openai>=0.2.0
+langchain-ollama>=0.2.0
+langchain-community>=0.3.0
+
+# 向量数据库
+chromadb>=0.5.0
+
+# 验证服务
+fastapi>=0.115.0
+uvicorn[standard]>=0.30.0
+lxml>=5.3.0
+
+# 嵌入模型（本地）
+sentence-transformers>=3.0.0
+
+# 工具类
+python-dotenv>=1.0.0
+httpx>=0.27.0
+tiktoken>=0.7.0
+
+# 测试
+pytest>=8.0.0
+pytest-asyncio>=0.24.0