nonebot-plugin-llm-qa-system 0.1.3__py3-none-any.whl

nonebot_plugin_llm_qa_system/__init__.py

@@ -0,0 +1,354 @@
+"""nonebot_plugin_llm_qa_system - 基于 RAG 的智能问答系统
+
+基于本地 Ollama 大模型 + 语义检索的知识问答机器人。
+
+命令:
+  问答 <问题>          — 基于知识库回答用户问题
+  添加知识 <标题> <内容>  — 向知识库添加条目
+  删除知识 <id>         — 删除指定知识条目
+  列出知识              — 列出知识库所有条目
+  清空知识              — 清空知识库（需确认）
+  搜索知识 <关键词>      — 搜索知识库
+"""
+
+from __future__ import annotations
+
+import json
+from typing import Any
+
+from nonebot import on_command, logger, require
+from nonebot.adapters.onebot.v11 import Bot, GroupMessageEvent, permission as perm
+from nonebot.params import CommandArg
+from nonebot.permission import SUPERUSER
+from nonebot.plugin import PluginMetadata
+
+require("nonebot_plugin_orm")
+
+from nonebot_plugin_orm import get_session as get_orm_session
+from sqlalchemy import delete, select
+
+from .config import Config
+from .models import KnowledgeEntry
+from .rag_engine import RAGEngine
+
+__plugin_meta__ = PluginMetadata(
+    name="nonebot-plugin-llm-qa-system",
+    description="基于 Ollama + RAG 的智能问答系统",
+    usage=(
+        "问答 <问题> — 基于知识库回答\n"
+        "添加知识 <标题> <内容> — 添加知识条目\n"
+        "删除知识 <id> — 删除指定条目\n"
+        "列出知识 — 列出所有条目\n"
+        "搜索知识 <关键词> — 语义搜索\n"
+        "清空知识 — 清空全部（需确认）"
+    ),
+    type="application",
+    config=Config,
+)
+
+# ==================== 配置加载 ====================
+
+try:
+    from nonebot import get_plugin_config
+    plugin_config = get_plugin_config(Config)
+except ImportError:
+    from nonebot import get_driver
+    plugin_config = Config.parse_obj(get_driver().config)
+
+# ==================== 全局引擎 ====================
+
+_engine: RAGEngine | None = None
+
+
+async def _get_engine() -> RAGEngine:
+    """获取或初始化 RAG 引擎。"""
+    global _engine
+    if _engine is None:
+        _engine = RAGEngine(plugin_config)
+    return _engine
+
+
+# ==================== 问答命令 ====================
+
+qa_cmd = on_command("问答", permission=perm.GROUP, priority=10, block=True)
+
+
+@qa_cmd.handle()
+async def handle_qa(
+    bot: Bot,
+    event: GroupMessageEvent,
+    args: Any = CommandArg(),
+) -> None:
+    """基于知识库回答用户问题。"""
+    query = args.extract_plain_text().strip()
+    if not query:
+        await qa_cmd.finish("用法：问答 <你的问题>")
+
+    await qa_cmd.send(f"🔍 正在思考：{query}")
+
+    # 加载知识库
+    async with get_orm_session() as session:
+        stmt = select(KnowledgeEntry)
+        result = await session.execute(stmt)
+        entries = result.scalars().all()
+
+    if not entries:
+        await qa_cmd.finish("知识库为空，请先添加知识。\n用法：添加知识 <标题> <内容>")
+
+    # 转为 dict 供检索
+    entry_dicts = [
+        {
+            "id": e.id,
+            "title": e.title,
+            "content": e.content,
+            "embedding": e.embedding,
+        }
+        for e in entries
+    ]
+
+    engine = await _get_engine()
+
+    # 检索
+    await qa_cmd.send("📚 正在检索相关知识...")
+    try:
+        relevant = await engine.retrieve(query, entry_dicts)
+    except Exception as e:
+        logger.error(f"llm_qa: 检索失败: {e}")
+        await qa_cmd.finish(f"❌ 检索失败：{e}")
+        return
+
+    if not relevant:
+        await qa_cmd.finish("未找到相关问题，请尝试换一种问法。")
+
+    # 生成回答
+    await qa_cmd.send("🤖 正在生成回答...")
+    answer = await engine.ask(query, relevant)
+
+    # 构建回复
+    sources = [f"[{i+1}] {c.get('title', '未知')}" for i, c in enumerate(relevant)]
+    reply = (
+        f"💡 回答：\n{answer}\n\n"
+        f"📎 参考来源：\n" + "\n".join(sources)
+    )
+    await qa_cmd.finish(reply)
+
+
+# ==================== 知识管理命令 ====================
+
+add_cmd = on_command("添加知识", permission=perm.GROUP, priority=10, block=True)
+
+
+@add_cmd.handle()
+async def handle_add_knowledge(
+    bot: Bot,
+    event: GroupMessageEvent,
+    args: Any = CommandArg(),
+) -> None:
+    """添加知识条目。"""
+    text = args.extract_plain_text().strip()
+    parts = text.split(maxsplit=1)
+    if len(parts) < 2:
+        await add_cmd.finish("用法：添加知识 <标题> <内容>")
+
+    title = parts[0]
+    content = parts[1]
+
+    # 生成嵌入
+    engine = await _get_engine()
+    try:
+        embedding = await engine.embed(f"{title}\n{content}")
+    except Exception as e:
+        logger.error(f"llm_qa: 生成嵌入失败: {e}")
+        await add_cmd.finish(f"❌ 生成嵌入向量失败，无法添加知识：{e}")
+        return
+
+    if not embedding:
+        await add_cmd.finish("❌ 嵌入向量返回为空，请检查 Ollama 嵌入模型是否可用。")
+        return
+
+    # 入库
+    async with get_orm_session() as session:
+        entry = KnowledgeEntry(
+            title=title,
+            content=content,
+            embedding=json.dumps(embedding),
+        )
+        session.add(entry)
+        await session.flush()  # 先 flush 让数据库生成自增 ID
+        entry_id = entry.id    # flush 后 id 已填充到实例中，此时访问不会触发 lazy load
+        await session.commit()
+
+    await add_cmd.finish(
+        f"✅ 已添加知识 #{entry_id}\n"
+        f"标题：{title}\n"
+        f"内容：{content[:100]}{'...' if len(content) > 100 else ''}"
+    )
+
+
+# ==================== 列出知识 ====================
+
+list_cmd = on_command("列出知识", permission=perm.GROUP, priority=10, block=True)
+
+
+@list_cmd.handle()
+async def handle_list_knowledge(
+    bot: Bot,
+    event: GroupMessageEvent,
+) -> None:
+    """列出知识库所有条目。"""
+    async with get_orm_session() as session:
+        stmt = select(KnowledgeEntry).order_by(KnowledgeEntry.id)
+        result = await session.execute(stmt)
+        entries = result.scalars().all()
+
+    if not entries:
+        await list_cmd.finish("📭 知识库为空")
+
+    lines = ["📚 知识库列表："]
+    for e in entries:
+        preview = e.content[:80].replace("\n", " ")
+        lines.append(f"  #{e.id} {e.title} — {preview}{'...' if len(e.content) > 80 else ''}")
+    lines.append(f"\n共 {len(entries)} 条")
+
+    # 分批发送避免消息过长
+    msg = "\n".join(lines)
+    if len(msg) > 1500:
+        chunks = []
+        current = []
+        for line in lines:
+            if current and len("\n".join(current + [line])) > 1000:
+                chunks.append("\n".join(current))
+                current = [line]
+            else:
+                current.append(line)
+        if current:
+            chunks.append("\n".join(current))
+        # 除最后一条外都用 send，最后一条用 finish
+        for chunk in chunks[:-1]:
+            await list_cmd.send(chunk)
+        await list_cmd.finish(chunks[-1])
+    else:
+        await list_cmd.finish(msg)
+
+
+# ==================== 搜索知识 ====================
+
+search_cmd = on_command("搜索知识", permission=perm.GROUP, priority=10, block=True)
+
+
+@search_cmd.handle()
+async def handle_search_knowledge(
+    bot: Bot,
+    event: GroupMessageEvent,
+    args: Any = CommandArg(),
+) -> None:
+    """语义搜索知识库。"""
+    query = args.extract_plain_text().strip()
+    if not query:
+        await search_cmd.finish("用法：搜索知识 <关键词>")
+
+    async with get_orm_session() as session:
+        stmt = select(KnowledgeEntry)
+        result = await session.execute(stmt)
+        entries = result.scalars().all()
+
+    if not entries:
+        await search_cmd.finish("📭 知识库为空")
+
+    entry_dicts = [
+        {"id": e.id, "title": e.title, "content": e.content, "embedding": e.embedding}
+        for e in entries
+    ]
+
+    engine = await _get_engine()
+    try:
+        relevant = await engine.retrieve(query, entry_dicts, top_k=5)
+    except Exception as e:
+        logger.error(f"llm_qa: 搜索失败: {e}")
+        await search_cmd.finish(f"❌ 搜索失败：{e}")
+        return
+
+    if not relevant:
+        await search_cmd.finish(f"未找到与「{query}」相关的内容")
+
+    lines = [f"🔍 搜索「{query}」结果："]
+    for i, c in enumerate(relevant, 1):
+        content_preview = c["content"][:100].replace("\n", " ")
+        lines.append(f"  #{c['id']} [{i}] {c['title']}")
+        lines.append(f"     {content_preview}{'...' if len(c['content']) > 100 else ''}")
+
+    await search_cmd.finish("\n".join(lines))
+
+
+# ==================== 删除知识 ====================
+
+del_cmd = on_command("删除知识", permission=SUPERUSER, priority=10, block=True)
+
+
+@del_cmd.handle()
+async def handle_delete_knowledge(
+    bot: Bot,
+    event: GroupMessageEvent,
+    args: Any = CommandArg(),
+) -> None:
+    """删除指定知识条目。"""
+    text = args.extract_plain_text().strip()
+    if not text.isdigit():
+        await del_cmd.finish("用法：删除知识 <ID>")
+
+    entry_id = int(text)
+
+    async with get_orm_session() as session:
+        stmt = select(KnowledgeEntry).where(KnowledgeEntry.id == entry_id)
+        result = await session.execute(stmt)
+        entry = result.scalar_one_or_none()
+        if entry is None:
+            await del_cmd.finish(f"❌ 未找到 ID 为 {entry_id} 的知识条目")
+
+        title = entry.title
+        await session.delete(entry)
+        await session.commit()
+
+    await del_cmd.finish(f"🗑️ 已删除 #{entry_id} {title}")
+
+
+# ==================== 清空知识 ====================
+
+clear_cmd = on_command("清空知识", permission=SUPERUSER, priority=10, block=True)
+
+
+@clear_cmd.handle()
+async def handle_clear_knowledge(
+    bot: Bot,
+    event: GroupMessageEvent,
+    args: Any = CommandArg(),
+) -> None:
+    """清空知识库。"""
+    confirm = args.extract_plain_text().strip()
+    if confirm != "确认":
+        await clear_cmd.finish(
+            "⚠️ 确认要清空所有知识条目吗？\n"
+            "此操作不可撤销。\n"
+            "请发送：清空知识 确认"
+        )
+
+    async with get_orm_session() as session:
+        stmt = delete(KnowledgeEntry)
+        result = await session.execute(stmt)
+        await session.commit()
+        count = result.rowcount
+
+    await clear_cmd.finish(f"🗑️ 已清空知识库，共删除 {count} 条")
+
+
+# ==================== 启动/关闭事件 ====================
+
+from nonebot import get_driver
+
+driver = get_driver()
+
+
+@driver.on_shutdown
+async def _():
+    if _engine is not None:
+        await _engine.close()

nonebot_plugin_llm_qa_system/config.py

@@ -0,0 +1,29 @@
+"""nonebot_plugin_llm_qa_system - 配置"""
+
+from pydantic import BaseModel, Extra
+
+
+class Config(BaseModel, extra=Extra.ignore):
+    """插件配置项，在 .env 文件中设置"""
+
+    # Ollama 服务地址
+    llm_qa_ollama_host: str = "http://localhost:11434"
+
+    # 对话模型名称
+    llm_qa_chat_model: str = "qwen3:1.7b"
+
+    # 嵌入模型名称
+    llm_qa_embed_model: str = "nomic-embed-text"
+
+    # RAG 检索返回的最大相关文档数
+    llm_qa_top_k: int = 3
+
+    # 余弦相似度最低阈值，低于该值的条目不返回也不展示
+    llm_qa_min_score: float = 0.3
+
+    # 系统提示词
+    llm_qa_system_prompt: str = (
+        "你是一个智能问答助手。请根据提供的参考信息，"
+        "用中文回答用户的问题。如果参考信息不足以回答问题，"
+        "请如实告知，不要编造答案。"
+    )

nonebot_plugin_llm_qa_system/models.py

@@ -0,0 +1,20 @@
+"""nonebot_plugin_llm_qa_system - ORM 数据模型"""
+
+from nonebot import require
+
+require("nonebot_plugin_orm")
+
+from nonebot_plugin_orm import Model
+from sqlalchemy import TEXT, Integer, String
+from sqlalchemy.orm import Mapped, mapped_column
+
+
+class KnowledgeEntry(Model):
+    """知识条目表"""
+
+    __tablename__ = "llm_qa_knowledge"
+
+    id: Mapped[int] = mapped_column(primary_key=True, autoincrement=True)
+    title: Mapped[str] = mapped_column(String(255), comment="标题/关键词")
+    content: Mapped[str] = mapped_column(TEXT, comment="知识内容")
+    embedding: Mapped[str] = mapped_column(TEXT, comment="嵌入向量（JSON数组）", default="[]")

nonebot_plugin_llm_qa_system/rag_engine.py

@@ -0,0 +1,210 @@
+"""nonebot_plugin_llm_qa_system - RAG 引擎（Ollama 嵌入 + 语义搜索 + LLM 生成）"""
+
+import json
+import math
+from typing import Any
+
+import httpx
+from nonebot import logger
+
+from .config import Config
+
+
+class RAGEngine:
+    """基于 Ollama 的 RAG 引擎，提供嵌入、检索、问答能力。"""
+
+    def __init__(self, config: Config) -> None:
+        self.config = config
+        self._http = httpx.AsyncClient(
+            base_url=config.llm_qa_ollama_host,
+            timeout=60,
+        )
+        self._embed_api_ver: int | None = None  # 1 = /api/embeddings, 2 = /api/embed
+
+    # ==================== 嵌入 ====================
+
+    async def embed(self, text: str) -> list[float]:
+        """调用 Ollama 生成文本嵌入向量。
+
+        优先尝试新版 /api/embed API，失败时回退到旧版 /api/embeddings。
+        """
+        if self._embed_api_ver == 2 or self._embed_api_ver is None:
+            try:
+                return await self._embed_v2(text)
+            except Exception as e:
+                if self._embed_api_ver == 2:
+                    raise
+                logger.warning(f"llm_qa: /api/embed 失败，尝试 /api/embeddings: {e}")
+
+        return await self._embed_v1(text)
+
+    async def _embed_v2(self, text: str) -> list[float]:
+        """新版 Ollama 嵌入 API (>=0.1.24)"""
+        resp = await self._http.post(
+            "/api/embed",
+            json={
+                "model": self.config.llm_qa_embed_model,
+                "input": text,
+            },
+        )
+        resp.raise_for_status()
+        data = resp.json()
+
+        # 新版返回 embeddings: list[list[float]]
+        embeddings = data.get("embeddings")
+        if embeddings and isinstance(embeddings, list) and len(embeddings) > 0:
+            self._embed_api_ver = 2
+            return embeddings[0]
+
+        # 某些版本可能返回 embedding: list[float]
+        single = data.get("embedding")
+        if single and isinstance(single, list):
+            self._embed_api_ver = 2
+            return single
+
+        raise RuntimeError(f"无法解析 /api/embed 响应: {data.keys()}")
+
+    async def _embed_v1(self, text: str) -> list[float]:
+        """旧版 Ollama 嵌入 API"""
+        resp = await self._http.post(
+            "/api/embeddings",
+            json={
+                "model": self.config.llm_qa_embed_model,
+                "prompt": text,
+            },
+        )
+        resp.raise_for_status()
+        data = resp.json()
+
+        embedding = data.get("embedding")
+        if embedding and isinstance(embedding, list):
+            self._embed_api_ver = 1
+            return embedding
+
+        raise RuntimeError(f"无法解析 /api/embeddings 响应: {data.keys()}")
+
+    # ==================== 相似度 ====================
+
+    @staticmethod
+    def cosine_similarity(a: list[float], b: list[float]) -> float:
+        """计算两个向量的余弦相似度。"""
+        if not a or not b or len(a) != len(b):
+            return 0.0
+        dot = sum(x * y for x, y in zip(a, b))
+        norm_a = math.sqrt(sum(x * x for x in a))
+        norm_b = math.sqrt(sum(x * x for x in b))
+        if norm_a == 0 or norm_b == 0:
+            return 0.0
+        return dot / (norm_a * norm_b)
+
+    # ==================== 检索 ====================
+
+    async def retrieve(
+        self,
+        query: str,
+        entries: list[dict[str, Any]],
+        top_k: int | None = None,
+    ) -> list[dict[str, Any]]:
+        """检索与查询最相关的知识条目。
+
+        Args:
+            query: 用户查询文本。
+            entries: 知识条目列表，每项含 id, title, content, embedding。
+            top_k: 返回条数，默认使用配置值。
+
+        Returns:
+            按相似度降序排列的条目列表。
+
+        Raises:
+            RuntimeError: 嵌入生成失败时抛出。
+        """
+        if not entries:
+            return []
+
+        top_k = top_k or self.config.llm_qa_top_k
+        query_emb = await self.embed(query)
+        if not query_emb:
+            raise RuntimeError("查询嵌入生成失败，无法执行检索")
+
+        scored: list[tuple[float, dict[str, Any]]] = []
+        for entry in entries:
+            emb = json.loads(entry.get("embedding", "[]") or "[]")
+            if not emb:
+                continue
+            score = self.cosine_similarity(query_emb, emb)
+            scored.append((score, entry))
+
+        scored.sort(key=lambda x: -x[0])
+        min_score = self.config.llm_qa_min_score
+        return [entry for score, entry in scored[:top_k] if score >= min_score]
+
+    # ==================== 问答 ====================
+
+    async def ask(
+        self,
+        query: str,
+        context_chunks: list[dict[str, Any]],
+        max_context_chars: int = 6000,
+    ) -> str:
+        """调用 Ollama 生成回答。
+
+        Args:
+            query: 用户问题。
+            context_chunks: 检索到的相关条目。
+            max_context_chars: 上下文最大字符数，超出时逐个截断条目内容。
+
+        Returns:
+            LLM 生成的回答文本。
+        """
+        # 构建上下文文本（带长度限制）
+        context_parts: list[str] = []
+        current_len = 0
+        for i, chunk in enumerate(context_chunks, 1):
+            title = chunk.get("title", f"文档{i}")
+            content = chunk.get("content", "")
+            part = f"[{i}] {title}\n{content}"
+
+            remaining = max_context_chars - current_len
+            if remaining <= 0:
+                break
+            if len(part) > remaining:
+                part = part[:max(remaining - 30, 0)] + "\n...[内容过长，已截断]"
+            context_parts.append(part)
+            current_len += len(part)
+
+        context_text = "\n\n".join(context_parts)
+
+        messages = [
+            {"role": "system", "content": self.config.llm_qa_system_prompt},
+        ]
+        if context_text:
+            messages.append({
+                "role": "user",
+                "content": (
+                    f"请根据以下参考信息回答问题。\n\n"
+                    f"参考信息：\n{context_text}\n\n"
+                    f"问题：{query}"
+                ),
+            })
+        else:
+            messages.append({"role": "user", "content": query})
+
+        try:
+            resp = await self._http.post(
+                "/api/chat",
+                json={
+                    "model": self.config.llm_qa_chat_model,
+                    "messages": messages,
+                    "stream": False,
+                },
+            )
+            resp.raise_for_status()
+            data = resp.json()
+            return data.get("message", {}).get("content", "抱歉，我没有得到有效的回答。")
+        except Exception as e:
+            logger.error(f"llm_qa: LLM 调用失败: {e}")
+            return f"抱歉，调用语言模型时出错：{e}"
+
+    async def close(self) -> None:
+        """关闭 HTTP 客户端。"""
+        await self._http.aclose()

nonebot_plugin_llm_qa_system-0.1.3.dist-info/METADATA

@@ -0,0 +1,145 @@
+Metadata-Version: 2.4
+Name: nonebot-plugin-llm-qa-system
+Version: 0.1.3
+Summary: 基于 Ollama + RAG 的智能问答系统
+License: MIT
+License-File: LICENSE
+Keywords: nonebot,nonebot2,ollama,rag,qa
+Author: BG4JEC
+Author-email: BG4JEC@hotmail.com
+Requires-Python: >=3.9
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.9
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Programming Language :: Python :: 3.14
+Requires-Dist: httpx (>=0.24.0)
+Requires-Dist: nonebot-adapter-onebot (>=2.0.0,<3.0.0)
+Requires-Dist: nonebot-plugin-orm (>=1.0.0)
+Requires-Dist: nonebot2 (>=2.0.0,<3.0.0)
+Project-URL: Repository, https://github.com/2580m/nonebot-plugin-llm-qa-system
+Description-Content-Type: text/markdown
+
+# nonebot-plugin-llm-qa-system
+
+基于本地 Ollama 大模型 + RAG（检索增强生成）的 NoneBot2 智能问答插件。
+
+## 功能
+
+- **问答**：基于知识库的内容，利用 LLM 生成回答
+- **添加知识**：向知识库添加条目，自动生成语义嵌入向量
+- **语义搜索**：通过余弦相似度检索相关知识
+- **知识管理**：列出、删除、清空知识条目
+
+## 安装
+
+```bash
+pip install nonebot-plugin-llm-qa-system
+```
+
+或者将本插件目录复制到项目的 `src/plugins/` 下，然后在 `pyproject.toml` 中注册：
+
+```toml
+[tool.nonebot]
+plugins = ["nonebot_plugin_llm_qa_system"]
+```
+
+## 前置依赖
+
+- [Ollama](https://ollama.com/) 本地运行
+- 所需的模型（首次使用前需拉取）：
+
+```bash
+ollama pull qwen3:1.7b         # 对话模型（默认）
+ollama pull nomic-embed-text   # 嵌入模型
+```
+
+## 配置
+
+在项目 `.env` 文件中添加以下配置项：
+
+```env
+# —— 数据库（必须）——
+# 默认使用 SQLite，通过 nonebot-plugin-orm 管理
+# 可自定义路径，确保目录已创建
+SQLALCHEMY_DATABASE_URL=sqlite+aiosqlite:///path/to/data/llm_qa.db
+
+# —— 插件配置 ——
+# Ollama 服务地址（默认值 http://localhost:11434）
+llm_qa_ollama_host=http://localhost:11434
+
+# 对话模型名称（默认值 qwen3:1.7b）
+llm_qa_chat_model=qwen3:1.7b
+
+# 嵌入模型名称（默认值 nomic-embed-text）
+llm_qa_embed_model=nomic-embed-text
+
+# RAG 检索返回的最大相关文档数（默认值 3）
+llm_qa_top_k=3
+
+# 余弦相似度最低阈值，低于该值的结果不返回（默认值 0.3）
+llm_qa_min_score=0.3
+```
+
+## 使用
+
+插件目前仅支持 **QQ 群聊**，所有命令通过群消息触发。
+
+| 命令 | 权限 | 说明 |
+|------|------|------|
+| `问答 <问题>` | 群员 | 基于知识库回答用户问题 |
+| `添加知识 <标题> <内容>` | 群员 | 向知识库添加条目 |
+| `删除知识 <id>` | SUPERUSER | 删除指定条目 |
+| `列出知识` | 群员 | 列出知识库所有条目 |
+| `搜索知识 <关键词>` | 群员 | 语义搜索知识库 |
+| `清空知识` | SUPERUSER | 清空全部条目（需确认） |
+
+### 示例
+
+```
+问答 RHEL 是什么？
+添加知识 Docker安装 使用以下命令安装 Docker...
+删除知识 3
+列出知识
+搜索知识 防火墙配置
+清空知识 确认
+```
+
+## 工作原理
+
+```
+用户提问 → 嵌入查询向量 → 余弦相似度检索知识库 → 拼接上下文 → LLM 生成回答
+                ↓                    ↑
+          Ollama nomic-embed-text    知识库（SQLite + SQLAlchemy ORM）
+                                     ↓
+                               Ollama qwen3:1.7b
+```
+
+1. 用户发送 `问答 <问题>`
+2. 插件从 SQLite 加载全部知识条目
+3. 调用 Ollama 的嵌入 API 将问题转为向量
+4. 计算所有条目的余弦相似度，返回 top_k 中高于 min_score 的条目
+5. 拼接为 Prompt 发送给 Ollama 对话模型
+6. 返回 LLM 生成的回答和参考来源
+
+## 兼容性
+
+插件自动兼容不同版本的 Ollama 嵌入 API：
+- 优先尝试新版 `/api/embed`（Ollama >= 0.1.24）
+- 失败时自动降级到旧版 `/api/embeddings`
+
+## 依赖
+
+- `nonebot2>=2.0.0`
+- `nonebot-adapter-onebot>=2.0.0`
+- `nonebot-plugin-orm>=1.0.0`
+- `httpx>=0.24.0`
+- `Ollama`（外部服务）
+
+## 许可证
+
+MIT
+

nonebot_plugin_llm_qa_system-0.1.3.dist-info/RECORD

@@ -0,0 +1,8 @@
+nonebot_plugin_llm_qa_system/__init__.py,sha256=5QC3ZBn1lT2Zwxov1K50Nwe33uy-OznhbnQrvAuDUDQ,10676
+nonebot_plugin_llm_qa_system/config.py,sha256=5MwqhNqyzLjymOW0wLxqgcA1sRAu7J2vlg6DQj11RPY,872
+nonebot_plugin_llm_qa_system/models.py,sha256=jz4_yNzBBgXUgI0WFKA0aoU9DZRau2I3tKieWJed6Oo,672
+nonebot_plugin_llm_qa_system/rag_engine.py,sha256=8cgpIvvxlLwsQWlCAMHsWjwBGORTekuaWQpwYrTTzWc,7031
+nonebot_plugin_llm_qa_system-0.1.3.dist-info/licenses/LICENSE,sha256=FyJNmoVZFPTaveI-LXwvas62zHdjc9fiH6D84Y9_rF4,1083
+nonebot_plugin_llm_qa_system-0.1.3.dist-info/METADATA,sha256=pEMLKUYLIEdOfy2-WcULpecDZ8_eGDC3IDAUg_kVjEg,4310
+nonebot_plugin_llm_qa_system-0.1.3.dist-info/WHEEL,sha256=EGEvSphFYqXKs23-kQBeyNoJP1nrT8ZJKQoi5p5DYL8,88
+nonebot_plugin_llm_qa_system-0.1.3.dist-info/RECORD,,

nonebot_plugin_llm_qa_system-0.1.3.dist-info/WHEEL

@@ -0,0 +1,4 @@
+Wheel-Version: 1.0
+Generator: poetry-core 2.4.0
+Root-Is-Purelib: true
+Tag: py3-none-any

nonebot_plugin_llm_qa_system-0.1.3.dist-info/licenses/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 BG4JEC
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.