kris/ForcePilot
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

Merge branch 'main' of https://github.com/xerrors/Yuxi-Know into main

Browse Source
This commit is contained in:
Wenjie Zhang 2025-12-03 21:15:46 +08:00
commit d80bb13b7e
10 changed files with 156 additions and 69 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
docs/latest/changelog/roadmap.md
View File

@ -15,12 +15,12 @@
- 同名文件处理逻辑：遇到同名文件则在上传区域提示，是否删除旧文件
- conversation 待修改为异步的版本
- DBManager 需要将数据库修改为异步的aiosqlite或者异步mysql，缓存使用Redis存储
- agent 状态中的文件区域，新增可以下载
### Bugs
- 部分异常状态下，智能体的模型名称出现重叠[#279](https://github.com/xerrors/Yuxi-Know/issues/279)
- DeepSeek 官方接口适配会出现问题
- 目前的知识库的图片存在公开访问风险
- 深度分析智能体需要考虑上下文超限的问题
### 新增
- 优化知识库详情页面，更加简洁清晰
@ -34,6 +34,8 @@
- 新增自定义模型支持、新增 dashscope rerank/embeddings 模型的支持
- 新增文档解析的图片支持，已支持 MinerU Officical、Docs、Markdown Zip格式
- 新增暗色模式支持并调整整体 UI（[#343](https://github.com/xerrors/Yuxi-Know/pull/343)）
- agent 状态中的文件区域，新增可以下载
- 移除 Chroma 的支持，当前版本标记为移除
### 修复
- 修复重排序模型实际未生效的问题

10
server/routers/knowledge_router.py
View File

@ -87,10 +87,12 @@ async def create_database(
"""创建知识库"""
logger.debug(
f"Create database {database_name} with kb_type {kb_type}, "
f"additional_params {additional_params}, llm_info {llm_info}"
f"additional_params {additional_params}, llm_info {llm_info}, "
f"embed_model_name {embed_model_name}"
)
try:
additional_params = {**(additional_params or {})}
additional_params["auto_generate_questions"] = False # 默认不生成问题
def normalize_reranker_config(kb: str, params: dict) -> None:
reranker_cfg = params.get("reranker_config")
@ -112,12 +114,12 @@ async def create_database(
if not isinstance(reranker_cfg, Mapping):
raise HTTPException(status_code=400, detail="reranker_config must be an object")
enabled = bool(reranker_cfg.get("enabled", False))
reranker_enabled = bool(reranker_cfg.get("enabled", False))
model = (reranker_cfg.get("model") or "").strip()
recall_top_k = max(1, int(reranker_cfg.get("recall_top_k", 50)))
final_top_k = max(1, int(reranker_cfg.get("final_top_k", 10)))
if enabled:
if reranker_enabled:
if not model:
raise HTTPException(status_code=400, detail="reranker_config.model is required when enabled")
if model not in config.reranker_names:
@ -132,7 +134,7 @@ async def create_database(
model = model if model in config.reranker_names else ""
params["reranker_config"] = {
"enabled": enabled,
"enabled": reranker_enabled,
"model": model,
"recall_top_k": recall_top_k,
"final_top_k": final_top_k,

1
src/agents/common/mcp_repos/arxiv-mcp-server Submodule

@ -0,0 +1 @@
Subproject commit 057e2000be7b56823239815b0fe7c7fc0dbced96

1
src/agents/common/mcp_repos/mcp-server-mysql Submodule

@ -0,0 +1 @@
Subproject commit 6a0367834ea0fb5e5c94b9711e3e2756966789ea

9
src/config/static/models.py
View File

@ -29,7 +29,7 @@ class EmbedModelInfo(BaseModel):
dimension: int = Field(..., description="向量维度")
base_url: str = Field(..., description="API 基础 URL")
api_key: str = Field(..., description="API Key 或环境变量名")
model_id: str | None = Field(None, description="可选的模型 ID")
class RerankerInfo(BaseModel):
"""重排序模型配置"""
@ -158,42 +158,49 @@ DEFAULT_CHAT_MODEL_PROVIDERS: dict[str, ChatModelProvider] = {
DEFAULT_EMBED_MODELS: dict[str, EmbedModelInfo] = {
"siliconflow/BAAI/bge-m3": EmbedModelInfo(
model_id="siliconflow/BAAI/bge-m3",
name="BAAI/bge-m3",
dimension=1024,
base_url="https://api.siliconflow.cn/v1/embeddings",
api_key="SILICONFLOW_API_KEY",
),
"siliconflow/Pro/BAAI/bge-m3": EmbedModelInfo(
model_id="siliconflow/Pro/BAAI/bge-m3",
name="Pro/BAAI/bge-m3",
dimension=1024,
base_url="https://api.siliconflow.cn/v1/embeddings",
api_key="SILICONFLOW_API_KEY",
),
"siliconflow/Qwen/Qwen3-Embedding-0.6B": EmbedModelInfo(
model_id="siliconflow/Qwen/Qwen3-Embedding-0.6B",
name="Qwen/Qwen3-Embedding-0.6B",
dimension=1024,
base_url="https://api.siliconflow.cn/v1/embeddings",
api_key="SILICONFLOW_API_KEY",
),
"vllm/Qwen/Qwen3-Embedding-0.6B": EmbedModelInfo(
model_id="vllm/Qwen/Qwen3-Embedding-0.6B",
name="Qwen3-Embedding-0.6B",
dimension=1024,
base_url="http://localhost:8000/v1/embeddings",
api_key="no_api_key",
),
"ollama/nomic-embed-text": EmbedModelInfo(
model_id="ollama/nomic-embed-text",
name="nomic-embed-text",
dimension=768,
base_url="http://localhost:11434/api/embed",
api_key="no_api_key",
),
"ollama/bge-m3": EmbedModelInfo(
model_id="ollama/bge-m3",
name="bge-m3",
dimension=1024,
base_url="http://localhost:11434/api/embed",
api_key="no_api_key",
),
"dashscope/text-embedding-v4": EmbedModelInfo(
model_id="dashscope/text-embedding-v4",
name="text-embedding-v4",
dimension=1024,
base_url="https://dashscope.aliyuncs.com/compatible-mode/v1/embeddings",

35
src/knowledge/implementations/milvus.py
View File

@ -7,6 +7,7 @@ from typing import Any
from pymilvus import Collection, CollectionSchema, DataType, FieldSchema, connections, db, utility
from src import config
from src.knowledge.base import KnowledgeBase
from src.knowledge.indexing import process_file_to_markdown
from src.knowledge.utils.kb_utils import (
@ -91,10 +92,14 @@ class MilvusKB(KnowledgeBase):
"""创建 Milvus 集合"""
logger.info(f"Creating Milvus collection for {db_id}")
if db_id not in self.databases_meta:
if not (metadata := self.databases_meta.get(db_id)):
raise ValueError(f"Database {db_id} not found")
embed_info = self.databases_meta[db_id].get("embed_info", {})
# embed_info = metadata.get("embed_info", {})
if not (embed_info := metadata.get("embed_info")):
logger.error(f"Embedding info not found for database {db_id}, using default model")
embed_info = config.embed_model_names[config.embed_model]
collection_name = db_id
try:
@ -117,8 +122,8 @@ class MilvusKB(KnowledgeBase):
except Exception:
# 创建新集合
embedding_dim = getattr(embed_info, "dimension", 1024) if embed_info else 1024
model_name = getattr(embed_info, "name", "default") if embed_info else "default"
embedding_dim = embed_info.get("dimension", 1024)
model_name = embed_info.get("name", "default")
# 定义集合Schema
fields = [
@ -142,7 +147,7 @@ class MilvusKB(KnowledgeBase):
index_params = {"metric_type": "COSINE", "index_type": "IVF_FLAT", "params": {"nlist": 1024}}
collection.create_index("embedding", index_params)
logger.info(f"Created new Milvus collection: {collection_name}")
logger.info(f"Created new Milvus collection: {collection_name}: {model_name=}, {embedding_dim=}")
return collection
@ -154,25 +159,29 @@ class MilvusKB(KnowledgeBase):
except Exception as e:
logger.warning(f"Failed to load collection into memory: {e}")
def _get_async_embedding_function(self, embed_info: dict):
def _get_async_embedding(self, embed_info: dict):
"""获取 embedding 函数"""
# 检查是否有 model_id 字段，优先使用 select_embedding_model
if embed_info and "model_id" in embed_info:
from src.models.embed import select_embedding_model
return select_embedding_model(embed_info["model_id"])
# 使用原有的逻辑（兼容模式））
config_dict = get_embedding_config(embed_info)
embedding_model = OtherEmbedding(
return OtherEmbedding(
model=config_dict.get("model"),
base_url=config_dict.get("base_url"),
api_key=config_dict.get("api_key"),
)
def _get_async_embedding_function(self, embed_info: dict):
"""获取 embedding 函数"""
embedding_model = self._get_async_embedding(embed_info)
return partial(embedding_model.abatch_encode, batch_size=40)
def _get_embedding_function(self, embed_info: dict):
"""获取 embedding 函数"""
config_dict = get_embedding_config(embed_info)
embedding_model = OtherEmbedding(
model=config_dict.get("model"),
base_url=config_dict.get("base_url"),
api_key=config_dict.get("api_key"),
)
embedding_model = self._get_async_embedding(embed_info)
return partial(embedding_model.batch_encode, batch_size=40)

6
src/knowledge/manager.py
View File

@ -246,16 +246,12 @@ class KnowledgeBaseManager:
db_id = db_info["db_id"]
async with self._metadata_lock:
# 准备 additional_params，包含 auto_generate_questions
saved_params = kwargs.copy()
saved_params["auto_generate_questions"] = False
self.global_databases_meta[db_id] = {
"name": database_name,
"description": description,
"kb_type": kb_type,
"created_at": utc_isoformat(),
"additional_params": saved_params,
"additional_params": kwargs.copy(),
}
self._save_global_metadata()

14
src/knowledge/utils/kb_utils.py
View File

@ -247,15 +247,23 @@ def get_embedding_config(embed_info: dict) -> dict:
try:
if embed_info:
# 处理 embed_info 可能是字典或 EmbedModelInfo 对象的情况
if hasattr(embed_info, "name"):
# 优先检查是否有 model_id 字段
if "model_id" in embed_info:
from src.models.embed import select_embedding_model
model = select_embedding_model(embed_info["model_id"])
config_dict["model"] = model.model
config_dict["api_key"] = model.api_key
config_dict["base_url"] = model.base_url
config_dict["dimension"] = getattr(model, "dimension", 1024)
elif hasattr(embed_info, "name"):
# EmbedModelInfo 对象
config_dict["model"] = embed_info.name
config_dict["api_key"] = os.getenv(embed_info.api_key) or embed_info.api_key
config_dict["base_url"] = embed_info.base_url
config_dict["dimension"] = embed_info.dimension
else:
# 字典形式
# 字典形式（保持向后兼容）
config_dict["model"] = embed_info["name"]
config_dict["api_key"] = os.getenv(embed_info["api_key"]) or embed_info["api_key"]
config_dict["base_url"] = embed_info["base_url"]

3
src/models/embed.py
View File

@ -11,7 +11,7 @@ from src.utils import get_docker_safe_url, hashstr, logger
class BaseEmbeddingModel(ABC):
def __init__(self, model=None, name=None, dimension=None, url=None, base_url=None, api_key=None):
def __init__(self, model=None, name=None, dimension=None, url=None, base_url=None, api_key=None, model_id=None):
"""
Args:
model: 模型名称，冗余设计，同name
@ -140,6 +140,7 @@ class OllamaEmbedding(BaseEmbeddingModel):
payload = {"model": self.model, "input": message}
async with httpx.AsyncClient() as client:
try:
print(f"\n\n\nOllama Embedding request: {payload}\n\n\n")
response = await client.post(self.base_url, json=payload, timeout=60)
response.raise_for_status()
result = response.json()

142
web/src/views/DataBaseView.vue
View File

@ -14,15 +14,23 @@
<h3>知识库类型<span style="color: var(--color-error-500)">*</span></h3>
<div class="kb-type-cards">
<div
v-for="(typeInfo, typeKey) in supportedKbTypes"
v-for="(typeInfo, typeKey) in orderedKbTypes"
:key="typeKey"
class="kb-type-card"
:class="{ active: newDatabase.kb_type === typeKey }"
:data-type="typeKey"
@click="handleKbTypeChange(typeKey)"
>
<div class="card-header">
<component :is="getKbTypeIcon(typeKey)" class="type-icon" />
<span class="type-title">{{ getKbTypeLabel(typeKey) }}</span>
<a-tooltip
v-if="typeKey === 'chroma'"
title="Chroma 已标记为弃用状态，建议使用 Milvus 替代。同时会在下个正式版本中移除。"
placement="top"
>
<span class="deprecated-badge">弃用</span>
</a-tooltip>
</div>
<div class="card-description">{{ typeInfo.description }}</div>
</div>
@ -71,7 +79,7 @@
<a-textarea
v-model:value="newDatabase.description"
placeholder="新建知识库描述"
:auto-size="{ minRows: 5, maxRows: 10 }"
:auto-size="{ minRows: 3, maxRows: 10 }"
/>
<h3 style="margin-top: 20px;">隐私设置</h3>
@ -158,6 +166,18 @@
<p>正在加载知识库...</p>
</div>
<!-- 空状态显示 -->
<div v-else-if="!databases || databases.length === 0" class="empty-state">
<h3 class="empty-title">暂无知识库</h3>
<p class="empty-description">创建您的第一个知识库，开始管理文档和知识</p>
<a-button type="primary" size="large" @click="state.openNewDatabaseModel = true">
<template #icon>
<PlusOutlined />
</template>
创建知识库
</a-button>
</div>
<!-- 数据库列表 -->
<div v-else class="databases">
<div
@ -213,7 +233,7 @@ import { useRouter, useRoute } from 'vue-router';
import { useConfigStore } from '@/stores/config';
import { message } from 'ant-design-vue'
import { Database, Zap, FileDigit, Waypoints, Building2 } from 'lucide-vue-next';
import { LockOutlined, InfoCircleOutlined, QuestionCircleOutlined } from '@ant-design/icons-vue';
import { LockOutlined, InfoCircleOutlined, QuestionCircleOutlined, PlusOutlined } from '@ant-design/icons-vue';
import { databaseApi, typeApi } from '@/apis/knowledge_api';
import HeaderComponent from '@/components/HeaderComponent.vue';
import ModelSelectorComponent from '@/components/ModelSelectorComponent.vue';
@ -256,7 +276,7 @@ const createEmptyDatabaseForm = () => ({
name: '',
description: '',
embed_model_name: configStore.config?.embed_model,
kb_type: 'chroma',
kb_type: 'milvus',
is_private: false,
storage: '',
language: 'English',
@ -295,6 +315,27 @@ const llmModelSpec = computed(() => {
// 支持的知识库类型
const supportedKbTypes = ref({})
// 有序的知识库类型（Chroma 排在最后）
const orderedKbTypes = computed(() => {
const types = { ...supportedKbTypes.value }
const ordered = {}
const chromaData = types.chroma
// 先添加除了 Chroma 之外的所有类型
Object.keys(types).forEach(key => {
if (key !== 'chroma') {
ordered[key] = types[key]
}
})
// 最后添加 Chroma（如果存在）
if (chromaData) {
ordered.chroma = chromaData
}
return ordered
})
// 加载支持的知识库类型
const loadSupportedKbTypes = async () => {
try {
@ -368,24 +409,6 @@ const getKbTypeIcon = (type) => {
return icons[type] || Database
}
// const getKbTypeDescription = (type) => {
// const descriptions = {
// lightrag: '🔥 图结构索引 • 智能查询 • 关系挖掘 • 复杂推理',
// chroma: '⚡ 轻量向量 • 快速开发 • 本地部署 • 简单易用',
// milvus: '🚀 生产级 • 高性能 • 分布式 • 企业级部署'
// }
// return descriptions[type] || ''
// }
const getKbTypeAlertType = (type) => {
const types = {
lightrag: 'info',
chroma: 'success',
milvus: 'warning'
}
return types[type] || 'info'
}
const getKbTypeColor = (type) => {
const colors = {
lightrag: 'purple',
@ -395,7 +418,6 @@ const getKbTypeColor = (type) => {
return colors[type] || 'blue'
}
// 格式化创建时间
const formatCreatedTime = (createdAt) => {
if (!createdAt) return ''
@ -680,24 +702,12 @@ onMounted(() => {
border-color: var(--main-color);
}
// 为不同知识库类型设置不同的悬停颜色与主题色
&:nth-child(1):hover,
&:nth-child(1).active {
border-color: var(--color-accent-100);
.type-icon { color: var(--color-accent-500); }
&.active {
border-color: var(--main-color);
background: var(--main-10);
.type-icon { color: var(--main-color); }
}
&:nth-child(2):hover,
&:nth-child(2).active {
border-color: var(--color-warning-100);
.type-icon { color: var(--color-warning-500); }
}
&:nth-child(3):hover,
&:nth-child(3).active {
border-color: var(--color-error-100);
.type-icon { color: var(--color-error-500); }
}
.card-header {
display: flex;
align-items: center;
@ -725,6 +735,26 @@ onMounted(() => {
margin-bottom: 0;
// min-height: 40px;
}
.deprecated-badge {
background: var(--color-error-100);
color: var(--color-error-600);
font-size: 10px;
font-weight: 600;
padding: 2px 6px;
border-radius: 4px;
margin-left: auto;
text-transform: uppercase;
letter-spacing: 0.5px;
cursor: help;
transition: all 0.2s ease;
&:hover {
background: var(--color-error-200);
color: var(--color-error-700);
}
}
}
}
@ -900,8 +930,6 @@ onMounted(() => {
font-weight: 400;
flex: 1;
}
}
.database-empty {
@ -913,6 +941,38 @@ onMounted(() => {
color: var(--gray-900);
}
.empty-state {
display: flex;
flex-direction: column;
align-items: center;
justify-content: center;
padding: 100px 20px;
text-align: center;
.empty-title {
font-size: 20px;
font-weight: 600;
color: var(--gray-900);
margin: 0 0 12px 0;
letter-spacing: -0.02em;
}
.empty-description {
font-size: 14px;
color: var(--gray-600);
margin: 0 0 32px 0;
line-height: 1.5;
max-width: 320px;
}
.ant-btn {
height: 44px;
padding: 0 24px;
font-size: 15px;
font-weight: 500;
}
}
.database-container {
padding: 0;
}