feat: Enhance chat and knowledge base functionalities with multi-KB support

- Updated agentChat API to accept optional knowledge_base_ids and agent_enabled parameters.
- Introduced getSession API to retrieve session details.
- Modified useStream to validate knowledge_base_ids for agent-chat requests and prepare POST body accordingly.
- Enhanced createKnowledgeBase API to include additional configuration options.
- Refactored chat routing to remove knowledge base ID from the chat path, allowing for dynamic knowledge base selection.
- Improved session management by preventing duplicate session entries in the menu store.
- Updated UI components for better session handling and display.
- Added context management for agent conversations to improve state persistence across sessions.
This commit is contained in:
wizardchen
2025-11-06 21:12:26 +08:00
parent 602e8a3b88
commit 266b9a5943
41 changed files with 3188 additions and 1746 deletions
+15 -2
View File
@@ -19,8 +19,17 @@ export async function knowledgeChat(data: { session_id: string; query: string; }
}
// Agent chat with streaming support
export async function agentChat(data: { session_id: string; query: string; }) {
return postChat(`/api/v1/agent-chat/${data.session_id}`, { query: data.query });
export async function agentChat(data: {
session_id: string;
query: string;
knowledge_base_ids?: string[];
agent_enabled: boolean;
}) {
return postChat(`/api/v1/agent-chat/${data.session_id}`, {
query: data.query,
knowledge_base_ids: data.knowledge_base_ids,
agent_enabled: data.agent_enabled
});
}
export async function getMessageList(data: { session_id: string; limit: number, created_at: string }) {
@@ -33,4 +42,8 @@ export async function getMessageList(data: { session_id: string; limit: number,
export async function delSession(session_id: string) {
return del(`/api/v1/sessions/${session_id}`);
}
export async function getSession(session_id: string) {
return get(`/api/v1/sessions/${session_id}`);
}
+24 -2
View File
@@ -28,7 +28,7 @@ export function useStream() {
let renderTimer: number | null = null
// 启动流式请求
const startStream = async (params: { session_id: any; query: any; method: string; url: string }) => {
const startStream = async (params: { session_id: any; query: any; knowledge_base_ids?: string[]; agent_enabled?: boolean; method: string; url: string }) => {
// 重置状态
output.value = '';
error.value = null;
@@ -46,11 +46,33 @@ export function useStream() {
return;
}
// Validate knowledge_base_ids for agent-chat requests
const isAgentChat = params.url === '/api/v1/agent-chat';
if (isAgentChat && params.method === 'POST') {
if (!params.knowledge_base_ids || params.knowledge_base_ids.length === 0) {
error.value = "请至少选择一个知识库";
stopStream();
return;
}
}
try {
let url =
params.method == "POST"
? `${apiUrl}${params.url}/${params.session_id}`
: `${apiUrl}${params.url}/${params.session_id}?message_id=${params.query}`;
// Prepare POST body with required fields for agent-chat
// knowledge_base_ids array and agent_enabled can update Session's SessionAgentConfig
const postBody: any = {
query: params.query,
agent_enabled: params.agent_enabled !== undefined ? params.agent_enabled : true
};
// Always include knowledge_base_ids for agent-chat (already validated above)
if (params.knowledge_base_ids !== undefined && params.knowledge_base_ids.length > 0) {
postBody.knowledge_base_ids = params.knowledge_base_ids;
}
await fetchEventSource(url, {
method: params.method,
headers: {
@@ -60,7 +82,7 @@ export function useStream() {
},
body:
params.method == "POST"
? JSON.stringify({ query: params.query })
? JSON.stringify(postBody)
: null,
signal: controller.signal,
openWhenHidden: true,
+12 -1
View File
@@ -5,7 +5,18 @@ export function listKnowledgeBases() {
return get(`/api/v1/knowledge-bases`);
}
export function createKnowledgeBase(data: { name: string; description?: string; chunking_config?: any }) {
export function createKnowledgeBase(data: {
name: string;
description?: string;
chunking_config?: any;
embedding_model_id?: string;
summary_model_id?: string;
rerank_model_id?: string;
vlm_model_id?: string;
vlm_config?: any;
cos_config?: any;
extract_config?: any;
}) {
return post(`/api/v1/knowledge-bases`, data);
}
+193 -98
View File
@@ -49,32 +49,34 @@
</t-popup>
</div>
<div ref="submenuscrollContainer" @scroll="handleScroll" class="submenu" v-if="item.children">
<div class="submenu_item_p" v-for="(subitem, subindex) in item.children" :key="subindex"
@click="gotopage(subitem.path)">
<div :class="['submenu_item', currentSecondpath == subitem.path ? 'submenu_item_active' : '']"
@mouseenter="mouseenteBotDownr(subindex)" @mouseleave="mouseleaveBotDown">
<i v-if="currentSecondpath == subitem.path" class="dot"></i>
<span class="submenu_title"
:style="currentSecondpath == subitem.path ? 'margin-left:14px;max-width:160px;' : 'margin-left:18px;max-width:173px;'">
{{ subitem.title }}
</span>
<t-popup v-model:visible="subitem.isMore" @overlay-click="delCard(subindex, subitem)"
@visible-change="onVisibleChange" overlayClassName="del-menu-popup" trigger="click"
destroy-on-close placement="top-left">
<div v-if="(activeSubmenu == subindex) || (currentSecondpath == subitem.path) || subitem.isMore"
@click.stop="openMore(subindex)" variant="outline" class="menu-more-wrap">
<t-icon name="ellipsis" class="menu-more" />
</div>
<template #content>
<span class="del_submenu">删除记录</span>
</template>
</t-popup>
<template v-for="(group, groupIndex) in groupedSessions" :key="groupIndex">
<div class="timeline_header">{{ group.label }}</div>
<div class="submenu_item_p" v-for="(subitem, subindex) in group.items" :key="subitem.id">
<div :class="['submenu_item', currentSecondpath == subitem.path ? 'submenu_item_active' : '']"
@mouseenter="mouseenteBotDownr(subitem.id)" @mouseleave="mouseleaveBotDown"
@click="gotopage(subitem.path)">
<i v-if="currentSecondpath == subitem.path" class="dot"></i>
<span class="submenu_title"
:style="currentSecondpath == subitem.path ? 'margin-left:14px;max-width:160px;' : 'margin-left:18px;max-width:173px;'">
{{ subitem.title }}
</span>
<t-dropdown
:options="[{ content: '删除记录', value: 'delete' }]"
@click="(data) => data.value === 'delete' && delCard(subitem.originalIndex, subitem)"
placement="bottom-right"
trigger="click">
<div @click.stop class="menu-more-wrap">
<t-icon name="ellipsis" class="menu-more" />
</div>
</t-dropdown>
</div>
</div>
</div>
</template>
</div>
</div>
</div>
<!-- 下半部分:用户菜单 -->
<div class="menu_bottom">
<UserMenu />
@@ -114,13 +116,12 @@ const totalPages = computed(() => Math.ceil(total.value / page_size.value));
const hasMore = computed(() => currentPage.value < totalPages.value);
type MenuItem = { title: string; icon: string; path: string; childrenPath?: string; children?: any[] };
const { menuArr } = storeToRefs(usemenuStore);
let activeSubmenu = ref<number>(-1);
let activeSubmenu = ref<string>('');
// 是否处于知识库详情页
// 是否处于知识库详情页(不包括全局聊天)
const isInKnowledgeBase = computed<boolean>(() => {
return route.name === 'knowledgeBaseDetail' ||
route.name === 'kbCreatChat' ||
route.name === 'chat' ||
route.name === 'knowledgeBaseSettings';
});
@@ -187,6 +188,70 @@ const initializedKnowledgeBases = computed(() => {
)
})
// 时间分组函数
const getTimeCategory = (dateStr: string): string => {
if (!dateStr) return '更早';
const date = new Date(dateStr);
const now = new Date();
const today = new Date(now.getFullYear(), now.getMonth(), now.getDate());
const yesterday = new Date(today.getTime() - 24 * 60 * 60 * 1000);
const sevenDaysAgo = new Date(today.getTime() - 7 * 24 * 60 * 60 * 1000);
const thirtyDaysAgo = new Date(today.getTime() - 30 * 24 * 60 * 60 * 1000);
const oneYearAgo = new Date(today.getTime() - 365 * 24 * 60 * 60 * 1000);
const sessionDate = new Date(date.getFullYear(), date.getMonth(), date.getDate());
if (sessionDate.getTime() >= today.getTime()) {
return '今天';
} else if (sessionDate.getTime() >= yesterday.getTime()) {
return '昨天';
} else if (date.getTime() >= sevenDaysAgo.getTime()) {
return '近7天';
} else if (date.getTime() >= thirtyDaysAgo.getTime()) {
return '近30天';
} else if (date.getTime() >= oneYearAgo.getTime()) {
return '近1年';
} else {
return '更早';
}
};
// 按时间分组Session列表
const groupedSessions = computed(() => {
const chatMenu = (menuArr.value as unknown as MenuItem[]).find((item: MenuItem) => item.path === 'creatChat');
if (!chatMenu || !chatMenu.children || chatMenu.children.length === 0) {
return [];
}
const groups: { [key: string]: any[] } = {
'今天': [],
'昨天': [],
'近7天': [],
'近30天': [],
'近1年': [],
'更早': []
};
// 将sessions按时间分组
(chatMenu.children as any[]).forEach((session: any, index: number) => {
const category = getTimeCategory(session.updated_at || session.created_at);
groups[category].push({
...session,
originalIndex: index
});
});
// 按顺序返回非空分组
const orderedLabels = ['今天', '昨天', '近7天', '近30天', '近1年', '更早'];
return orderedLabels
.filter(label => groups[label].length > 0)
.map(label => ({
label,
items: groups[label]
}));
});
// 动态更新知识库菜单项标题
const kbMenuItem = computed(() => {
const kbItem = topMenuItems.value.find(item => item.path === 'knowledge-bases')
@@ -282,27 +347,27 @@ const upload = async (e: any) => {
uploadInput.value.value = "";
}
}
const mouseenteBotDownr = (val: number) => {
const mouseenteBotDownr = (val: string) => {
activeSubmenu.value = val;
}
const mouseleaveBotDown = () => {
activeSubmenu.value = -1;
}
const onVisibleChange = (_e: any) => {
activeSubmenu.value = '';
}
const delCard = (index: number, item: any) => {
delSession(item.id).then((res: any) => {
if (res && (res as any).success) {
(menuArr.value as any[])[1]?.children?.splice(index, 1);
// 使用 originalIndex 找到正确的位置进行删除
const actualIndex = index !== undefined ? index :
(menuArr.value as any[])[1]?.children?.findIndex((s: any) => s.id === item.id);
if (actualIndex !== -1) {
(menuArr.value as any[])[1]?.children?.splice(actualIndex, 1);
}
if (item.id == route.params.chatid) {
// 删除当前会话后,跳转到当前知识库的创建聊天页面
const kbId = route.params.kbId;
if (kbId) {
router.push(`/platform/knowledge-bases/${kbId}/creatChat`);
} else {
router.push('/platform/knowledge-bases');
}
// 删除当前会话后,跳转到全局创建聊天页面
router.push('/platform/creatChat');
}
} else {
MessagePlugin.error("删除失败,请稍后再试!");
@@ -319,70 +384,89 @@ const debounce = (fn: (...args: any[]) => void, delay: number) => {
// 滚动处理
const checkScrollBottom = () => {
const container = submenuscrollContainer.value
if (!container) return
if (!container || !container[0]) return
const { scrollTop, scrollHeight, clientHeight } = container[0]
const isBottom = scrollHeight - (scrollTop + clientHeight) < 100 // 触底阈值
if (isBottom && hasMore.value) {
if (isBottom && hasMore.value && !loading.value) {
currentPage.value++;
getMessageList();
getMessageList(true);
}
}
const handleScroll = debounce(checkScrollBottom, 200)
const getMessageList = async () => {
// 仅在知识库内部显示对话列表
if (!isInKnowledgeBase.value) {
usemenuStore.clearMenuArr();
currentKbName.value = '';
return;
}
const getMessageList = async (isLoadMore = false) => {
// 全局显示对话列表(所有有"对话"入口的地方都显示)
let kbId = (route.params as any)?.kbId as string
// 新的路由格式:/platform/chat/:kbId/:chatid,直接从路由参数获取知识库ID
if (!kbId) {
usemenuStore.clearMenuArr();
// 如果在知识库内部,获取知识库名称和所有知识库列表
if (kbId && isInKnowledgeBase.value) {
try {
const [kbRes, allKbRes]: any[] = await Promise.all([
getKnowledgeBaseById(kbId),
listKnowledgeBases()
])
if (kbRes?.data?.name) {
currentKbName.value = kbRes.data.name
}
if (allKbRes?.data) {
allKnowledgeBases.value = allKbRes.data
}
} catch {}
} else {
// 不在知识库内部时,清空知识库名称
currentKbName.value = '';
return;
// 不在知识库列表页时才调用API(避免与页面组件KnowledgeBaseList.vue重复调用)
// 知识库列表页会自己调用listKnowledgeBases来显示列表
if (route.name !== 'knowledgeBaseList') {
try {
const allKbRes: any = await listKnowledgeBases()
if (allKbRes?.data) {
allKnowledgeBases.value = allKbRes.data
}
} catch {}
}
}
// 获取知识库名称和所有知识库列表
try {
const [kbRes, allKbRes]: any[] = await Promise.all([
getKnowledgeBaseById(kbId),
listKnowledgeBases()
])
if (kbRes?.data?.name) {
currentKbName.value = kbRes.data.name
}
if (allKbRes?.data) {
allKnowledgeBases.value = allKbRes.data
}
} catch {}
if (loading.value) return;
if (loading.value) return Promise.resolve();
loading.value = true;
usemenuStore.clearMenuArr();
getSessionsList(currentPage.value, page_size.value).then((res: any) => {
// 只有在首次加载或路由变化时才清空数组,滚动加载时不清空
if (!isLoadMore) {
currentPage.value = 1; // 重置页码
usemenuStore.clearMenuArr();
}
return getSessionsList(currentPage.value, page_size.value).then((res: any) => {
if (res.data && res.data.length) {
// 过滤出当前知识库的会话
const filtered = res.data.filter((s: any) => s.knowledge_base_id === kbId)
filtered.forEach((item: any) => {
let obj = { title: item.title ? item.title : "新会话", path: `chat/${kbId}/${item.id}`, id: item.id, isMore: false, isNoTitle: item.title ? false : true }
// Display all sessions globally without filtering
res.data.forEach((item: any) => {
let obj = {
title: item.title ? item.title : "新会话",
path: `chat/${item.id}`,
id: item.id,
isMore: false,
isNoTitle: item.title ? false : true,
created_at: item.created_at,
updated_at: item.updated_at
}
usemenuStore.updatemenuArr(obj)
});
loading.value = false;
}
if ((res as any).total) {
total.value = (res as any).total;
}
loading.value = false;
}).catch(() => {
loading.value = false;
})
}
const openMore = (_e: any) => { }
onMounted(() => {
const routeName = typeof route.name === 'string' ? route.name : (route.name ? String(route.name) : '')
currentpath.value = routeName;
if (route.params.chatid && route.params.kbId) {
currentSecondpath.value = `chat/${route.params.kbId}/${route.params.chatid}`;
if (route.params.chatid) {
currentSecondpath.value = `chat/${route.params.chatid}`;
}
getMessageList();
});
@@ -390,12 +474,12 @@ onMounted(() => {
watch([() => route.name, () => route.params], (newvalue) => {
const nameStr = typeof newvalue[0] === 'string' ? (newvalue[0] as string) : (newvalue[0] ? String(newvalue[0]) : '')
currentpath.value = nameStr;
if (newvalue[1].chatid && newvalue[1].kbId) {
currentSecondpath.value = `chat/${newvalue[1].kbId}/${newvalue[1].chatid}`;
if (newvalue[1].chatid) {
currentSecondpath.value = `chat/${newvalue[1].chatid}`;
} else {
currentSecondpath.value = "";
}
// 路由变化时刷新对话列表(仅在知识库内部)
// 路由变化时刷新对话列表
getMessageList();
// 路由变化时更新图标状态
getIcon(nameStr);
@@ -553,11 +637,6 @@ watch(() => route.params.kbId, () => {
</script>
<style lang="less" scoped>
.del_submenu {
color: #fa5151;
cursor: pointer;
}
.aside_box {
min-width: 260px;
padding: 8px;
@@ -733,10 +812,25 @@ watch(() => route.params.kbId, () => {
min-height: 0;
margin-left: 4px;
}
.timeline_header {
font-family: "PingFang SC";
font-size: 12px;
font-weight: 600;
color: #00000066;
padding: 12px 18px 6px 18px;
margin-top: 8px;
line-height: 20px;
user-select: none;
&:first-child {
margin-top: 4px;
}
}
.submenu_item_p {
height: 44px;
padding: 4px 8px 4px 12px;
padding: 4px 8px 4px 0px;
box-sizing: border-box;
}
@@ -749,7 +843,7 @@ watch(() => route.params.kbId, () => {
font-weight: 400;
line-height: 22px;
height: 36px;
padding-left: 18px;
padding-left: 0px;
padding-right: 14px;
position: relative;
@@ -761,6 +855,8 @@ watch(() => route.params.kbId, () => {
.menu-more-wrap {
margin-left: auto;
opacity: 0;
transition: opacity 0.2s ease;
}
.menu-more {
@@ -789,6 +885,10 @@ watch(() => route.params.kbId, () => {
color: #00000099;
}
.menu-more-wrap {
opacity: 1;
}
.submenu_title {
max-width: 160px !important;
@@ -804,6 +904,14 @@ watch(() => route.params.kbId, () => {
.menu-more {
color: #07c05f !important;
}
.menu-more-wrap {
opacity: 1;
}
.submenu_title {
max-width: 160px !important;
}
}
}
@@ -908,19 +1016,6 @@ watch(() => route.params.kbId, () => {
}
}
.del-menu-popup {
z-index: 99 !important;
.t-popup__content {
width: 100px;
height: 40px;
line-height: 30px;
padding-left: 14px;
cursor: pointer;
margin-top: 4px !important;
}
}
// 退出登录确认框样式
:deep(.t-popconfirm) {
+1 -1
View File
@@ -71,7 +71,7 @@ const router = createRouter({
meta: { requiresInit: true, requiresAuth: true }
},
{
path: "chat/:kbId/:chatid",
path: "chat/:chatid",
name: "chat",
component: () => import("../views/chat/index.vue"),
meta: { requiresInit: true, requiresAuth: true }
+5 -1
View File
@@ -25,7 +25,11 @@ export const useMenuStore = defineStore('menuStore', {
this.menuArr[1].children = reactive<object[]>([]);
},
updatemenuArr(obj: any) {
this.menuArr[1].children?.push(obj);
// 检查是否已存在相同 ID 的 session,避免重复添加
const exists = this.menuArr[1].children?.some((item: any) => item.id === obj.id);
if (!exists) {
this.menuArr[1].children?.push(obj);
}
},
updataMenuChildren(item: object) {
this.menuArr[1].children?.unshift(item)
+12 -4
View File
@@ -3,7 +3,8 @@ import { defineStore } from 'pinia'
export const useUIStore = defineStore('ui', {
state: () => ({
showSettingsModal: false,
showKBSettingsModal: false,
showKBEditorModal: false,
kbEditorMode: 'create' as 'create' | 'edit',
currentKBId: null as string | null,
settingsInitialSection: null as string | null,
settingsInitialSubSection: null as string | null
@@ -28,11 +29,18 @@ export const useUIStore = defineStore('ui', {
openKBSettings(kbId: string) {
this.currentKBId = kbId
this.showKBSettingsModal = true
this.kbEditorMode = 'edit'
this.showKBEditorModal = true
},
closeKBSettings() {
this.showKBSettingsModal = false
openCreateKB() {
this.currentKBId = null
this.kbEditorMode = 'create'
this.showKBEditorModal = true
},
closeKBEditor() {
this.showKBEditorModal = false
this.currentKBId = null
}
}
@@ -17,7 +17,10 @@
<div v-if="event.type === 'thinking'" class="thinking-event">
<div
class="thinking-phase"
:class="{ 'thinking-active': event.thinking }"
:class="{
'thinking-active': event.thinking,
'thinking-last': isLastThinking(event.event_id)
}"
>
<div v-if="event.content"
class="thinking-content markdown-content"
@@ -136,6 +139,26 @@ const expandedEvents = ref<Set<string>>(new Set());
// State for intermediate steps collapse
const showIntermediateSteps = ref(false);
// Find the last thinking event in the current message's event stream
// Only the last thinking should have the green border-left
const lastThinkingEventId = computed(() => {
const stream = eventStream.value;
if (!stream || stream.length === 0) return null;
// Find all thinking events
const thinkingEvents = stream.filter((e: any) => e.type === 'thinking');
if (thinkingEvents.length === 0) return null;
// Return the event_id of the last thinking event
const lastThinking = thinkingEvents[thinkingEvents.length - 1];
return lastThinking.event_id;
});
// Check if a thinking event is the last one (should have green border)
const isLastThinking = (eventId: string): boolean => {
return eventId === lastThinkingEventId.value;
};
// Check if conversation is done (based on answer event with done=true)
const isConversationDone = computed(() => {
const stream = eventStream.value;
@@ -498,10 +521,18 @@ const formatJSON = (obj: any): string => {
background: #ffffff;
border-radius: 8px;
padding: 1px 14px;
border-left: 3px solid #d1d5db;
// 默认不显示 border-left
border-left: 3px solid transparent;
box-shadow: 0 1px 2px rgba(0, 0, 0, 0.04);
transition: all 0.25s cubic-bezier(0.4, 0, 0.2, 1);
// 只有最后一个 thinking 显示绿色 border-left
&.thinking-last {
border-left-color: #07c05f;
box-shadow: 0 2px 4px rgba(7, 192, 95, 0.08);
}
// 进行中的 thinking(实时对话)显示绿色 border-left 和动画
&.thinking-active {
border-left-color: #07c05f;
box-shadow: 0 2px 4px rgba(7, 192, 95, 0.08);
@@ -5,7 +5,7 @@
<AgentStreamDisplay :session="session" v-if="session.isAgentMode"></AgentStreamDisplay>
<deepThink :deepSession="session" v-if="session.showThink && !session.isAgentMode"></deepThink>
</div>
<div ref="parentMd">
<div ref="parentMd" v-if="!session.hideContent">
<!-- 消息正在总结中则渲染加载gif -->
<img v-if="session.thinking" class="botanswer_laoding_gif" src="@/assets/img/botanswer_loading.gif"
alt="正在总结答案……">
@@ -212,11 +212,11 @@ onMounted(async () => {
}
.bot_msg {
background: #fff;
// background: #fff;
border-radius: 4px;
color: rgba(0, 0, 0, 0.9);
font-size: 16px;
padding: 10px 12px;
// padding: 10px 12px;
margin-right: auto;
max-width: 100%;
box-sizing: border-box;
+140 -38
View File
@@ -29,10 +29,11 @@ import { useRoute, useRouter, onBeforeRouteLeave, onBeforeRouteUpdate } from 'vu
import InputField from '../../components/Input-field.vue';
import botmsg from './components/botmsg.vue';
import usermsg from './components/usermsg.vue';
import { getMessageList, generateSessionsTitle } from "@/api/chat/index";
import { getMessageList, generateSessionsTitle, getSession } from "@/api/chat/index";
import { useStream } from '../../api/chat/streame'
import { useMenuStore } from '@/stores/menu';
import { useSettingsStore } from '@/stores/settings';
import { MessagePlugin } from 'tdesign-vue-next';
const usemenuStore = useMenuStore();
const useSettingsStoreInstance = useSettingsStore();
const { menuArr, isFirstSession, firstQuery } = storeToRefs(usemenuStore);
@@ -40,7 +41,7 @@ const { output, onChunk, isStreaming, isLoading, error, startStream, stopStream
const route = useRoute();
const router = useRouter();
const session_id = ref(route.params.chatid);
const knowledge_base_id = ref(route.params.kbId);
const sessionData = ref(null);
const created_at = ref('');
const limit = ref(20);
const messagesList = reactive([]);
@@ -60,7 +61,6 @@ watch([() => route.params], (newvalue) => {
}
messagesList.splice(0);
session_id.value = newvalue[0].chatid;
knowledge_base_id.value = newvalue[0].kbId;
checkmenuTitle(session_id.value)
let data = {
session_id: session_id.value,
@@ -107,6 +107,59 @@ const getmsgList = (data, isScrollType = false, scrollHeight) => {
}
})
}
// Reconstruct agentEventStream from agent_steps stored in database
// This allows the frontend to restore the exact conversation state including all agent reasoning steps
const reconstructEventStreamFromSteps = (agentSteps) => {
if (!agentSteps || !Array.isArray(agentSteps) || agentSteps.length === 0) {
return [];
}
const events = [];
agentSteps.forEach((step) => {
// Add thinking event if thought content exists
if (step.thought && step.thought.trim()) {
events.push({
type: 'thinking',
event_id: `step-${step.iteration}-thought`,
content: step.thought,
done: true,
thinking: false,
});
}
// Add tool call and result events
if (step.tool_calls && Array.isArray(step.tool_calls)) {
step.tool_calls.forEach((toolCall) => {
events.push({
type: 'tool_call',
tool_call_id: toolCall.id,
tool_name: toolCall.name,
arguments: toolCall.args,
pending: false,
success: toolCall.result?.success !== false,
output: toolCall.result?.output || '',
error: toolCall.result?.error || undefined,
duration: toolCall.duration,
display_type: toolCall.result?.data?.display_type,
tool_data: toolCall.result?.data,
});
});
}
});
// 添加一个完成标记的 answer 事件,触发折叠逻辑
if (events.length > 0) {
events.push({
type: 'answer',
content: '', // 空内容,因为最终答案会在 message.content 中
done: true
});
}
return events;
};
const handleMsgList = async (data, isScrollType = false, newScrollHeight) => {
let chatlist = data.reverse()
for (let i = 0, len = chatlist.length; i < len; i++) {
@@ -116,6 +169,17 @@ const handleMsgList = async (data, isScrollType = false, newScrollHeight) => {
item._eventMap = new Map();
item._pendingToolCalls = new Map();
// Check if this message has agent_steps from database (historical agent conversation)
// If so, reconstruct the agentEventStream to restore the exact conversation state
if (item.agent_steps && Array.isArray(item.agent_steps) && item.agent_steps.length > 0) {
console.log('[Message Load] Reconstructing agent steps for message:', item.id, 'steps:', item.agent_steps.length);
item.isAgentMode = true;
item.agentEventStream = reconstructEventStreamFromSteps(item.agent_steps);
// 隐藏最终答案内容,用户可以通过展开步骤查看完整过程
item.hideContent = true;
console.log('[Message Load] Reconstructed', item.agentEventStream.length, 'events from agent steps');
}
if (item.content) {
if (!item.content.includes('<think>') && !item.content.includes('<\/think>')) {
item.thinkContent = "";
@@ -163,19 +227,42 @@ const sendMsg = async (value) => {
messagesList.push({ content: value, role: 'user' });
scrollToBottom();
// 判断是否使用 Agent 模式
const isAgentMode = useSettingsStoreInstance.isAgentEnabled;
const chatUrl = isAgentMode ? '/api/v1/agent-chat' : '/api/v1/knowledge-chat';
// Always use agent mode with unified architecture
// Get knowledge_base_ids from session's agent_config.knowledge_bases to update SessionAgentConfig
let kbIds = [];
if (sessionData.value?.agent_config?.knowledge_bases?.length > 0) {
kbIds = sessionData.value.agent_config.knowledge_bases;
}
// Validate knowledge_base_ids before sending
if (kbIds.length === 0) {
MessagePlugin.warning('请至少选择一个知识库');
isReplying.value = false;
loading.value = false;
// Remove the user message that was just added
messagesList.pop();
return;
}
await startStream({
session_id: session_id.value,
knowledge_base_id: knowledge_base_id.value,
knowledge_base_ids: kbIds,
agent_enabled: true,
query: value,
method: 'POST',
url: chatUrl
url: '/api/v1/agent-chat'
});
}
// Watch for stream errors and show message
watch(error, (newError) => {
if (newError) {
MessagePlugin.error(newError);
isReplying.value = false;
loading.value = false;
}
});
// 处理流式数据
onChunk((data) => {
loading.value = false;
@@ -190,6 +277,33 @@ onChunk((data) => {
data: data.data
});
// 处理 agent query 事件 - 确保显示 loading 状态
if (data.response_type === 'agent_query') {
console.log('[Agent Query Event]', {
session_id: data.data?.session_id,
query: data.data?.query,
request_id: data.data?.request_id
});
// Loading state is already set in sendMsg, but ensure it's true
loading.value = true;
return;
}
// 处理会话标题更新事件
if (data.response_type === 'session_title') {
const title = data.content || data.data?.title;
if (title && data.data?.session_id) {
console.log('[Session Title Update]', {
session_id: data.data.session_id,
title: title
});
usemenuStore.updatasessionTitle(data.data.session_id, title);
usemenuStore.changeIsFirstSession(false);
isNeedTitle.value = false;
}
return;
}
// 判断是否是 Agent 模式的响应
const isAgentResponse = data.response_type === 'thinking' ||
data.response_type === 'tool_call' ||
@@ -220,17 +334,8 @@ onChunk((data) => {
obj.content = fullContent.value;
}
if (data.done) {
if (isFirstSession.value || isNeedTitle.value) {
generateSessionsTitle(session_id.value, {
messages: [{ role: "user", content: userquery.value }]
}).then(res => {
if (res.data) {
usemenuStore.changeIsFirstSession(false);
usemenuStore.updatasessionTitle(session_id.value, res.data);
isNeedTitle.value = false;
}
})
}
// 标题生成已改为异步事件推送,不再需要在这里手动调用
// 如果标题还未生成,前端会通过 SSE 事件接收
isReplying.value = false;
fullContent.value = "";
}
@@ -444,18 +549,8 @@ const handleAgentChunk = (data) => {
isReplying.value = false;
fullContent.value = '';
// 生成标题
if (isFirstSession.value || isNeedTitle.value) {
generateSessionsTitle(session_id.value, {
messages: [{ role: "user", content: userquery.value }]
}).then(res => {
if (res.data) {
usemenuStore.changeIsFirstSession(false);
usemenuStore.updatasessionTitle(session_id.value, res.data);
isNeedTitle.value = false;
}
});
}
// 标题生成已改为异步事件推送,不再需要在这里手动调用
// 如果标题还未生成,前端会通过 SSE 事件接收
}
break;
}
@@ -481,13 +576,19 @@ const updateAssistantSession = (payload) => {
}
scrollToBottom();
}
onMounted(() => {
onMounted(async () => {
messagesList.splice(0);
// scrollContainer.value.addEventListener("scroll", () => {
// if (scrollContainer.value.scrollTop == 0) {
// onChatScrollTop();
// }
// });
// Load session data to get agent_config
try {
const sessionRes = await getSession(session_id.value);
if (sessionRes?.data) {
sessionData.value = sessionRes.data;
}
} catch (error) {
console.error('Failed to load session data:', error);
}
checkmenuTitle(session_id.value)
if (firstQuery.value) {
scrollLock.value = true;
@@ -546,7 +647,6 @@ onBeforeRouteUpdate((to, from, next) => {
flex: 1;
width: 100%;
overflow-y: auto;
max-width: 800px;
&::-webkit-scrollbar {
width: 0;
@@ -586,6 +686,8 @@ onBeforeRouteUpdate((to, from, next) => {
gap: 16px;
max-width: 800px;
flex: 1;
margin: 0 auto;
width: 100%;
.botanswer_laoding_gif {
width: 24px;
+8 -5
View File
@@ -50,7 +50,7 @@ async function createNewSession(value: string) {
try {
const res = await createSessions(sessionData);
if (res.data && res.data.id) {
await navigateToSession(res.data.id, value, selectedKbs[0]);
await navigateToSession(res.data.id, value);
} else {
console.error("创建会话失败");
MessagePlugin.error("创建会话失败");
@@ -61,18 +61,21 @@ async function createNewSession(value: string) {
}
}
const navigateToSession = async (sessionId: string, value: string, kbId: string) => {
const navigateToSession = async (sessionId: string, value: string) => {
const now = new Date().toISOString();
let obj = {
title: '新会话',
path: `chat/${kbId}/${sessionId}`,
path: `chat/${sessionId}`,
id: sessionId,
isMore: false,
isNoTitle: true
isNoTitle: true,
created_at: now,
updated_at: now
};
usemenuStore.updataMenuChildren(obj);
usemenuStore.changeIsFirstSession(true);
usemenuStore.changeFirstQuery(value);
router.push(`/platform/chat/${kbId}/${sessionId}`);
router.push(`/platform/chat/${sessionId}`);
}
</script>
+10 -1
View File
@@ -176,7 +176,16 @@ const sendMsg = (value: string) => {
};
const getTitle = (session_id: string, value: string) => {
let obj = { title: '新会话', path: `chat/${kbId.value}/${session_id}`, id: session_id, isMore: false, isNoTitle: true };
const now = new Date().toISOString();
let obj = {
title: '新会话',
path: `chat/${kbId.value}/${session_id}`,
id: session_id,
isMore: false,
isNoTitle: true,
created_at: now,
updated_at: now
};
usemenuStore.updataMenuChildren(obj);
usemenuStore.changeIsFirstSession(true);
usemenuStore.changeFirstQuery(value);
@@ -0,0 +1,742 @@
<template>
<Teleport to="body">
<Transition name="modal">
<div v-if="visible" class="settings-overlay" @click.self="handleClose">
<div class="settings-modal">
<!-- 关闭按钮 -->
<button class="close-btn" @click="handleClose" aria-label="关闭">
<svg width="20" height="20" viewBox="0 0 20 20" fill="currentColor">
<path d="M15 5L5 15M5 5L15 15" stroke="currentColor" stroke-width="2" stroke-linecap="round"/>
</svg>
</button>
<div class="settings-container">
<!-- 左侧导航 -->
<div class="settings-sidebar">
<div class="sidebar-header">
<h2 class="sidebar-title">{{ mode === 'create' ? '新建知识库' : '知识库设置' }}</h2>
</div>
<div class="settings-nav">
<div
v-for="(item, index) in navItems"
:key="index"
:class="['nav-item', { 'active': currentSection === item.key }]"
@click="currentSection = item.key"
>
<span class="nav-icon">{{ item.icon }}</span>
<span class="nav-label">{{ item.label }}</span>
</div>
</div>
</div>
<!-- 右侧内容区域 -->
<div class="settings-content">
<div class="content-wrapper">
<!-- 基本信息 -->
<div v-show="currentSection === 'basic'" class="section">
<div v-if="formData" class="section-content">
<div class="section-header">
<h3 class="section-title">基本信息</h3>
<p class="section-desc">设置知识库的名称和描述信息</p>
</div>
<div class="section-body">
<div class="form-item">
<label class="form-label required">知识库名称</label>
<t-input
v-model="formData.name"
placeholder="请输入知识库名称"
:maxlength="50"
/>
</div>
<div class="form-item">
<label class="form-label">知识库描述</label>
<t-textarea
v-model="formData.description"
placeholder="请输入知识库描述(可选)"
:maxlength="200"
:autosize="{ minRows: 3, maxRows: 6 }"
/>
</div>
</div>
</div>
</div>
<!-- 模型配置 -->
<div v-show="currentSection === 'models'" class="section">
<KBModelConfig
ref="modelConfigRef"
v-if="formData"
:config="formData.modelConfig"
:has-files="hasFiles"
:all-models="allModels"
@update:config="handleModelConfigUpdate"
/>
</div>
<!-- 分块设置 -->
<div v-show="currentSection === 'chunking'" class="section">
<KBChunkingSettings
v-if="formData"
:config="formData.chunkingConfig"
@update:config="handleChunkingConfigUpdate"
/>
</div>
<!-- 高级设置 -->
<div v-show="currentSection === 'advanced'" class="section">
<KBAdvancedSettings
ref="advancedSettingsRef"
v-if="formData"
:multimodal="formData.multimodalConfig"
:node-extract="formData.nodeExtractConfig"
:all-models="allModels"
@update:multimodal="handleMultimodalUpdate"
@update:nodeExtract="handleNodeExtractUpdate"
/>
</div>
</div>
<!-- 保存按钮 -->
<div class="settings-footer">
<t-button theme="default" variant="outline" @click="handleClose">
取消
</t-button>
<t-button theme="primary" @click="handleSubmit" :loading="saving">
{{ mode === 'create' ? '创建知识库' : '保存配置' }}
</t-button>
</div>
</div>
</div>
</div>
</div>
</Transition>
</Teleport>
</template>
<script setup lang="ts">
import { ref, computed, watch } from 'vue'
import { MessagePlugin } from 'tdesign-vue-next'
import { createKnowledgeBase, getKnowledgeBaseById, listKnowledgeFiles, updateKnowledgeBase } from '@/api/knowledge-base'
import { updateKBConfig, type KBModelConfigRequest } from '@/api/initialization'
import { listModels } from '@/api/model'
import KBModelConfig from './settings/KBModelConfig.vue'
import KBChunkingSettings from './settings/KBChunkingSettings.vue'
import KBAdvancedSettings from './settings/KBAdvancedSettings.vue'
// Props
const props = defineProps<{
visible: boolean
mode: 'create' | 'edit'
kbId?: string
}>()
// Emits
const emit = defineEmits<{
(e: 'update:visible', value: boolean): void
(e: 'success', kbId: string): void
}>()
const currentSection = ref<string>('basic')
const saving = ref(false)
const loading = ref(false)
const allModels = ref<any[]>([])
const hasFiles = ref(false)
const navItems = [
{ key: 'basic', icon: 'ℹ️', label: '基本信息' },
{ key: 'models', icon: '🤖', label: '模型配置' },
{ key: 'chunking', icon: '📄', label: '分块设置' },
{ key: 'advanced', icon: '⚙️', label: '高级设置' }
]
// 模型配置引用
const modelConfigRef = ref<InstanceType<typeof KBModelConfig>>()
const advancedSettingsRef = ref<InstanceType<typeof KBAdvancedSettings>>()
// 表单数据
const formData = ref<any>(null)
// 初始化表单数据
const initFormData = () => {
return {
name: '',
description: '',
modelConfig: {
llmModelId: '',
embeddingModelId: '',
rerankModelId: ''
},
chunkingConfig: {
chunkSize: 512,
chunkOverlap: 100,
separators: ['\n\n', '\n', '。', '!', '?', ';', ';']
},
multimodalConfig: {
enabled: false,
storageType: 'minio' as 'minio' | 'cos',
vllmModelId: '',
minio: {
bucketName: '',
useSSL: false,
pathPrefix: ''
},
cos: {
secretId: '',
secretKey: '',
region: '',
bucketName: '',
appId: '',
pathPrefix: ''
}
},
nodeExtractConfig: {
enabled: false,
text: '',
tags: [] as string[]
}
}
}
// 加载所有模型
const loadAllModels = async () => {
try {
const models = await listModels()
allModels.value = models || []
} catch (error) {
console.error('加载模型列表失败:', error)
MessagePlugin.error('加载模型列表失败')
allModels.value = []
}
}
// 加载知识库数据(编辑模式)
const loadKBData = async () => {
if (props.mode !== 'edit' || !props.kbId) return
loading.value = true
try {
const [kbInfo, models, filesResult] = await Promise.all([
getKnowledgeBaseById(props.kbId),
loadAllModels(),
listKnowledgeFiles(props.kbId, { page: 1, page_size: 1 })
])
if (!kbInfo || !kbInfo.data) {
throw new Error('知识库不存在')
}
const kb = kbInfo.data
hasFiles.value = (filesResult as any)?.total > 0
// 设置表单数据
formData.value = {
name: kb.name || '',
description: kb.description || '',
modelConfig: {
llmModelId: kb.summary_model_id || '',
embeddingModelId: kb.embedding_model_id || '',
rerankModelId: kb.rerank_model_id || ''
},
chunkingConfig: {
chunkSize: kb.chunking_config?.chunk_size || 512,
chunkOverlap: kb.chunking_config?.chunk_overlap || 100,
separators: kb.chunking_config?.separators || ['\n\n', '\n', '。', '!', '?', ';', ';']
},
multimodalConfig: {
enabled: !!(kb.vlm_model_id || (kb.cos_config?.provider && kb.cos_config?.bucket_name)),
storageType: (kb.cos_config?.provider || 'minio') as 'minio' | 'cos',
vllmModelId: kb.vlm_model_id || '',
minio: {
bucketName: kb.cos_config?.bucket_name || '',
useSSL: kb.cos_config?.use_ssl || false,
pathPrefix: kb.cos_config?.path_prefix || ''
},
cos: {
secretId: kb.cos_config?.secret_id || '',
secretKey: kb.cos_config?.secret_key || '',
region: kb.cos_config?.region || '',
bucketName: kb.cos_config?.bucket_name || '',
appId: kb.cos_config?.app_id || '',
pathPrefix: kb.cos_config?.path_prefix || ''
}
},
nodeExtractConfig: {
enabled: kb.extract_config?.enabled || false,
text: kb.extract_config?.text || '',
tags: kb.extract_config?.tags || []
}
}
} catch (error) {
console.error('加载知识库数据失败:', error)
MessagePlugin.error('加载知识库数据失败')
handleClose()
} finally {
loading.value = false
}
}
// 处理配置更新
const handleModelConfigUpdate = (config: any) => {
if (formData.value) {
formData.value.modelConfig = { ...config }
}
}
const handleChunkingConfigUpdate = (config: any) => {
if (formData.value) {
formData.value.chunkingConfig = { ...config }
}
}
const handleMultimodalUpdate = (config: any) => {
if (formData.value) {
formData.value.multimodalConfig = { ...config }
}
}
const handleNodeExtractUpdate = (config: any) => {
if (formData.value) {
formData.value.nodeExtractConfig = { ...config }
}
}
// 验证表单
const validateForm = (): boolean => {
if (!formData.value) return false
// 验证基本信息
if (!formData.value.name || !formData.value.name.trim()) {
MessagePlugin.warning('请输入知识库名称')
currentSection.value = 'basic'
return false
}
// 验证模型配置 - 必须配置 embedding 和 summary 模型
if (!formData.value.modelConfig.embeddingModelId) {
MessagePlugin.warning('请选择 Embedding 模型')
currentSection.value = 'models'
return false
}
if (!formData.value.modelConfig.llmModelId) {
MessagePlugin.warning('请选择 Summary 模型')
currentSection.value = 'models'
return false
}
// 验证多模态配置(如果启用)
if (formData.value.multimodalConfig.enabled) {
const validation = advancedSettingsRef.value?.validateMultimodal()
if (validation && !validation.valid) {
MessagePlugin.warning(validation.message || '多模态配置验证失败')
currentSection.value = 'advanced'
return false
}
}
return true
}
// 构建提交数据
const buildSubmitData = () => {
if (!formData.value) return null
const data: any = {
name: formData.value.name,
description: formData.value.description,
chunking_config: {
chunk_size: formData.value.chunkingConfig.chunkSize,
chunk_overlap: formData.value.chunkingConfig.chunkOverlap,
separators: formData.value.chunkingConfig.separators,
enable_multimodal: formData.value.multimodalConfig.enabled
},
embedding_model_id: formData.value.modelConfig.embeddingModelId,
summary_model_id: formData.value.modelConfig.llmModelId
}
// 可选的 Rerank 模型
if (formData.value.modelConfig.rerankModelId) {
data.rerank_model_id = formData.value.modelConfig.rerankModelId
}
// 添加多模态配置
if (formData.value.multimodalConfig.enabled) {
if (formData.value.multimodalConfig.vllmModelId) {
data.vlm_model_id = formData.value.multimodalConfig.vllmModelId
}
const storageType = formData.value.multimodalConfig.storageType
if (storageType === 'minio') {
data.cos_config = {
provider: 'minio',
bucket_name: formData.value.multimodalConfig.minio.bucketName,
use_ssl: formData.value.multimodalConfig.minio.useSSL,
path_prefix: formData.value.multimodalConfig.minio.pathPrefix || undefined
}
} else {
data.cos_config = {
provider: 'cos',
secret_id: formData.value.multimodalConfig.cos.secretId,
secret_key: formData.value.multimodalConfig.cos.secretKey,
region: formData.value.multimodalConfig.cos.region,
bucket_name: formData.value.multimodalConfig.cos.bucketName,
app_id: formData.value.multimodalConfig.cos.appId,
path_prefix: formData.value.multimodalConfig.cos.pathPrefix || undefined
}
}
}
// 添加知识图谱配置
if (formData.value.nodeExtractConfig.enabled) {
data.extract_config = {
enabled: true,
text: formData.value.nodeExtractConfig.text,
tags: formData.value.nodeExtractConfig.tags
}
}
return data
}
// 提交表单
const handleSubmit = async () => {
if (!validateForm()) {
return
}
saving.value = true
try {
const data = buildSubmitData()
if (!data) {
throw new Error('数据构建失败')
}
if (props.mode === 'create') {
// 创建模式:一次性创建知识库及所有配置
const result = await createKnowledgeBase(data)
if (!result.success || !result.data?.id) {
throw new Error(result.message || '创建知识库失败')
}
MessagePlugin.success('知识库创建成功')
emit('success', result.data.id)
} else {
// 编辑模式:分别更新基本信息和配置
if (!props.kbId) {
throw new Error('缺少知识库ID')
}
// 1. 更新基本信息(名称、描述)
await updateKnowledgeBase(props.kbId, {
name: data.name,
description: data.description,
config: {} // 空配置,只更新基本信息
})
// 2. 更新完整配置(模型、分块、多模态、知识图谱等)
const config: KBModelConfigRequest = {
llmModelId: data.summary_model_id,
embeddingModelId: data.embedding_model_id,
rerankModelId: data.rerank_model_id || '',
vllmModelId: data.vlm_model_id || '',
documentSplitting: {
chunkSize: data.chunking_config.chunk_size,
chunkOverlap: data.chunking_config.chunk_overlap,
separators: data.chunking_config.separators
},
multimodal: {
enabled: !!data.cos_config,
storageType: data.cos_config?.provider || 'minio',
cos: data.cos_config?.provider === 'cos' ? {
secretId: data.cos_config.secret_id,
secretKey: data.cos_config.secret_key,
region: data.cos_config.region,
bucketName: data.cos_config.bucket_name,
appId: data.cos_config.app_id,
pathPrefix: data.cos_config.path_prefix || ''
} : undefined,
minio: data.cos_config?.provider === 'minio' ? {
bucketName: data.cos_config.bucket_name,
useSSL: data.cos_config.use_ssl || false,
pathPrefix: data.cos_config.path_prefix || ''
} : undefined
},
nodeExtract: {
enabled: data.extract_config?.enabled || false,
text: data.extract_config?.text || '',
tags: data.extract_config?.tags || [],
nodes: [],
relations: []
}
}
await updateKBConfig(props.kbId, config)
MessagePlugin.success('配置保存成功')
emit('success', props.kbId)
}
handleClose()
} catch (error: any) {
console.error('操作失败:', error)
MessagePlugin.error(error.message || '操作失败')
} finally {
saving.value = false
}
}
// 重置所有状态
const resetState = () => {
currentSection.value = 'basic'
formData.value = null
hasFiles.value = false
saving.value = false
loading.value = false
}
// 关闭弹窗
const handleClose = () => {
emit('update:visible', false)
setTimeout(() => {
resetState()
}, 300)
}
// 监听弹窗打开/关闭
watch(() => props.visible, async (newVal) => {
if (newVal) {
// 打开弹窗时,先重置状态
resetState()
// 加载模型列表
await loadAllModels()
// 根据模式加载数据
if (props.mode === 'edit' && props.kbId) {
await loadKBData()
} else {
// 创建模式:初始化空表单
formData.value = initFormData()
hasFiles.value = false
}
} else {
// 关闭弹窗时,延迟重置状态(等待动画结束)
setTimeout(() => {
resetState()
}, 300)
}
})
</script>
<style scoped lang="less">
// 复用创建知识库的样式
.settings-overlay {
position: fixed;
top: 0;
left: 0;
right: 0;
bottom: 0;
background: rgba(0, 0, 0, 0.5);
display: flex;
align-items: center;
justify-content: center;
z-index: 1000;
backdrop-filter: blur(4px);
}
.settings-modal {
position: relative;
width: 90vw;
max-width: 1100px;
height: 85vh;
max-height: 750px;
background: #fff;
border-radius: 12px;
box-shadow: 0 8px 32px rgba(0, 0, 0, 0.12);
display: flex;
flex-direction: column;
overflow: hidden;
}
.close-btn {
position: absolute;
top: 20px;
right: 20px;
width: 32px;
height: 32px;
border: none;
background: #f5f5f5;
border-radius: 6px;
cursor: pointer;
display: flex;
align-items: center;
justify-content: center;
color: #666;
transition: all 0.2s ease;
z-index: 10;
&:hover {
background: #e5e5e5;
color: #000;
}
}
.settings-container {
display: flex;
height: 100%;
overflow: hidden;
}
.settings-sidebar {
width: 200px;
background: #fafafa;
border-right: 1px solid #e5e5e5;
display: flex;
flex-direction: column;
flex-shrink: 0;
}
.sidebar-header {
padding: 24px 20px;
border-bottom: 1px solid #e5e5e5;
}
.sidebar-title {
margin: 0;
font-family: "PingFang SC";
font-size: 18px;
font-weight: 600;
color: #000000e6;
}
.settings-nav {
flex: 1;
padding: 12px 8px;
overflow-y: auto;
}
.nav-item {
display: flex;
align-items: center;
padding: 10px 12px;
margin-bottom: 4px;
border-radius: 6px;
cursor: pointer;
transition: all 0.2s ease;
font-family: "PingFang SC";
font-size: 14px;
color: #00000099;
&:hover {
background: #f0f0f0;
}
&.active {
background: #07c05f1a;
color: #07c05f;
font-weight: 500;
}
}
.nav-icon {
margin-right: 8px;
font-size: 16px;
}
.nav-label {
flex: 1;
}
.settings-content {
flex: 1;
display: flex;
flex-direction: column;
overflow: hidden;
}
.content-wrapper {
flex: 1;
overflow-y: auto;
padding: 24px 32px;
}
.section {
margin-bottom: 32px;
&:last-child {
margin-bottom: 0;
}
}
.section-content {
.section-header {
margin-bottom: 20px;
}
.section-title {
margin: 0 0 8px 0;
font-family: "PingFang SC";
font-size: 16px;
font-weight: 600;
color: #000000e6;
}
.section-desc {
margin: 0;
font-family: "PingFang SC";
font-size: 14px;
color: #00000066;
line-height: 22px;
}
.section-body {
background: #fff;
}
}
.form-item {
margin-bottom: 20px;
&:last-child {
margin-bottom: 0;
}
}
.form-label {
display: block;
margin-bottom: 8px;
font-family: "PingFang SC";
font-size: 14px;
font-weight: 500;
color: #000000e6;
&.required::after {
content: '*';
color: #FA5151;
margin-left: 4px;
}
}
.settings-footer {
padding: 16px 32px;
border-top: 1px solid #e5e5e5;
display: flex;
justify-content: flex-end;
gap: 12px;
flex-shrink: 0;
}
// 过渡动画
.modal-enter-active,
.modal-leave-active {
transition: all 0.3s ease;
}
.modal-enter-from,
.modal-leave-to {
opacity: 0;
.settings-modal {
transform: scale(0.95);
}
}
</style>
+541 -132
View File
@@ -1,8 +1,9 @@
<template>
<div class="kb-list-container">
<!-- 头部 -->
<div class="header">
<h2>知识库</h2>
<t-button theme="primary" @click="openCreate">新建知识库</t-button>
<button class="create-btn" @click="openCreateModal">新建知识库</button>
</div>
<!-- 未初始化知识库提示 -->
@@ -10,66 +11,107 @@
<t-icon name="info-circle" size="16px" />
<span>部分知识库尚未初始化,需要先在设置中配置模型信息才能添加知识文档</span>
</div>
<t-table :data="kbs" :columns="columns" row-key="id" size="medium" hover>
<template #status="{ row }">
<div class="status-cell">
<t-tag
:theme="isInitialized(row) ? 'success' : 'warning'"
size="small"
>
{{ isInitialized(row) ? '已初始化' : '未初始化' }}
</t-tag>
<t-tooltip
v-if="!isInitialized(row)"
content="需要先在设置中配置模型信息才能添加知识"
placement="top"
>
<span class="warning-icon">⚠</span>
</t-tooltip>
</div>
</template>
<template #description="{ row }">
<div class="description-text">{{ row.description || '暂无描述' }}</div>
</template>
<template #op="{ row }">
<t-space size="small">
<t-button
size="small"
@click="goDetail(row.id)"
:disabled="!isInitialized(row)"
:theme="isInitialized(row) ? 'primary' : 'default'"
:variant="isInitialized(row) ? 'base' : 'outline'"
:title="!isInitialized(row) ? '请先在设置中配置模型信息' : ''"
>
文档
</t-button>
<t-button size="small" variant="outline" @click="goSettings(row.id)">设置</t-button>
<t-popconfirm content="确认删除该知识库?" @confirm="remove(row.id)">
<t-button size="small" theme="danger" variant="text">删除</t-button>
</t-popconfirm>
</t-space>
</template>
</t-table>
<t-dialog v-model:visible="createVisible" header="新建知识库" :footer="false">
<t-form :data="createForm" @submit="create">
<t-form-item label="名称" name="name" :rules="[{ required: true, message: '请输入名称' }]">
<t-input v-model="createForm.name" />
</t-form-item>
<t-form-item label="描述" name="description">
<t-textarea v-model="createForm.description" />
</t-form-item>
<t-form-item>
<t-space>
<t-button theme="primary" type="submit" :loading="creating">创建</t-button>
<t-button variant="outline" @click="createVisible = false">取消</t-button>
</t-space>
</t-form-item>
</t-form>
<!-- 卡片网格 -->
<div v-if="kbs.length > 0" class="kb-card-wrap">
<div
v-for="(kb, index) in kbs"
:key="kb.id"
class="kb-card"
:class="{ 'uninitialized': !isInitialized(kb) }"
@click="handleCardClick(kb)"
>
<!-- 卡片头部 -->
<div class="card-header">
<span class="card-title" :title="kb.name">{{ kb.name }}</span>
<t-popup
v-model="kb.showMore"
overlayClassName="card-more-popup"
:on-visible-change="onVisibleChange"
trigger="click"
destroy-on-close
placement="bottom-right"
>
<div
variant="outline"
class="more-wrap"
@click.stop="openMore(index)"
:class="{ 'active-more': currentMoreIndex === index }"
>
<img class="more-icon" src="@/assets/img/more.png" alt="" />
</div>
<template #content>
<div class="popup-menu" @click.stop>
<div class="popup-menu-item" @click.stop="handleSettings(kb)">
<t-icon class="menu-icon" name="setting" />
<span>设置</span>
</div>
<div class="popup-menu-item delete" @click.stop="handleDelete(kb)">
<t-icon class="menu-icon" name="delete" />
<span>删除</span>
</div>
</div>
</template>
</t-popup>
</div>
<!-- 卡片内容 -->
<div class="card-content">
<div class="card-description">
{{ kb.description || '暂无描述' }}
</div>
</div>
<!-- 卡片底部 -->
<div class="card-bottom">
<div class="status-badge" :class="{ 'initialized': isInitialized(kb), 'uninitialized': !isInitialized(kb) }">
<span v-if="!isInitialized(kb)" class="warning-icon">⚠</span>
<span>{{ isInitialized(kb) ? '已初始化' : '未初始化' }}</span>
</div>
<span class="card-time">{{ kb.updated_at }}</span>
</div>
</div>
</div>
<!-- 空状态 -->
<div v-else-if="!loading" class="empty-state">
<img class="empty-img" src="@/assets/img/upload.svg" alt="">
<span class="empty-txt">暂无知识库</span>
<span class="empty-desc">点击右上角"新建知识库"按钮创建第一个知识库</span>
</div>
<!-- 删除确认对话框 -->
<t-dialog
v-model:visible="deleteVisible"
dialogClassName="del-knowledge-dialog"
:closeBtn="false"
:cancelBtn="null"
:confirmBtn="null"
>
<div class="circle-wrap">
<div class="dialog-header">
<img class="circle-img" src="@/assets/img/circle.png" alt="">
<span class="circle-title">删除确认</span>
</div>
<span class="del-circle-txt">
{{ `确认要删除知识库"${deletingKb?.name}"?删除后不可恢复` }}
</span>
<div class="circle-btn">
<span class="circle-btn-txt" @click="deleteVisible = false">取消</span>
<span class="circle-btn-txt confirm" @click="confirmDelete">确认删除</span>
</div>
</div>
</t-dialog>
<!-- 知识库设置模态框 -->
<KnowledgeBaseSettingsModal />
<!-- 知识库编辑器(创建/编辑统一组件) -->
<KnowledgeBaseEditorModal
:visible="uiStore.showKBEditorModal"
:mode="uiStore.kbEditorMode"
:kb-id="uiStore.currentKBId || undefined"
@update:visible="(val) => val ? null : uiStore.closeKBEditor()"
@success="handleKBEditorSuccess"
/>
<!-- 全局设置模态框 -->
<Settings />
@@ -77,13 +119,13 @@
</template>
<script setup lang="ts">
import { onMounted, reactive, ref, computed, watch } from 'vue'
import { onMounted, ref, computed, watch } from 'vue'
import { useRouter } from 'vue-router'
import { MessagePlugin } from 'tdesign-vue-next'
import { listKnowledgeBases, createKnowledgeBase, deleteKnowledgeBase } from '@/api/knowledge-base'
import { listKnowledgeBases, deleteKnowledgeBase } from '@/api/knowledge-base'
import { formatStringDate } from '@/utils/index'
import { useUIStore } from '@/stores/ui'
import KnowledgeBaseSettingsModal from './KnowledgeBaseSettingsModal.vue'
import KnowledgeBaseEditorModal from './KnowledgeBaseEditorModal.vue'
import Settings from '@/views/settings/Settings.vue'
const router = useRouter()
@@ -96,71 +138,78 @@ interface KB {
updated_at?: string;
embedding_model_id?: string;
summary_model_id?: string;
showMore?: boolean;
}
const kbs = ref<KB[]>([])
const loading = ref(false)
const columns = [
{ colKey: 'name', title: '名称' },
{ colKey: 'description', title: '描述', cell: 'description', width: 300 },
{ colKey: 'status', title: '状态', cell: 'status', width: 100 },
{ colKey: 'updated_at', title: '更新时间' },
{ colKey: 'op', title: '操作', cell: 'op', width: 220 },
]
const deleteVisible = ref(false)
const deletingKb = ref<KB | null>(null)
const currentMoreIndex = ref<number>(-1)
const fetchList = () => {
loading.value = true
listKnowledgeBases().then((res: any) => {
const data = res.data || []
// 格式化时间
// 格式化时间,并初始化 showMore 状态
kbs.value = data.map((kb: KB) => ({
...kb,
updated_at: kb.updated_at ? formatStringDate(new Date(kb.updated_at)) : ''
updated_at: kb.updated_at ? formatStringDate(new Date(kb.updated_at)) : '',
showMore: false
}))
}).finally(() => loading.value = false)
}
onMounted(fetchList)
onMounted(() => {
fetchList()
})
const createVisible = ref(false)
const creating = ref(false)
const createForm = reactive({ name: '', description: '' })
const openCreate = () => {
createForm.name = ''
createForm.description = ''
createVisible.value = true
// 打开创建知识库弹窗
const openCreateModal = () => {
uiStore.openCreateKB()
}
const create = () => {
if (!createForm.name) return
creating.value = true
const chunking_config = {
chunk_size: 512,
chunk_overlap: 100,
separators: ['.', '?', '!', '。', '?', '!'],
enable_multimodal: false
const openMore = (index: number) => {
// 只记录当前打开的索引,用于显示激活样式
// 弹窗的开关由 v-model 自动管理
currentMoreIndex.value = index
}
const onVisibleChange = (visible: boolean) => {
// 弹窗关闭时重置索引
if (!visible) {
currentMoreIndex.value = -1
}
createKnowledgeBase({ name: createForm.name, description: createForm.description, chunking_config }).then((res: any) => {
if (res.success) {
MessagePlugin.success('创建成功')
createVisible.value = false
fetchList()
} else {
MessagePlugin.error(res.message || '创建失败')
}
}).catch((e: any) => {
MessagePlugin.error(e?.message || '创建失败')
}).finally(() => creating.value = false)
}
const remove = (id: string) => {
deleteKnowledgeBase(id).then((res: any) => {
const handleSettings = (kb: KB) => {
// 手动关闭弹窗
kb.showMore = false
goSettings(kb.id)
}
const handleDelete = (kb: KB) => {
// 手动关闭弹窗
kb.showMore = false
deletingKb.value = kb
deleteVisible.value = true
}
const confirmDelete = () => {
if (!deletingKb.value) return
deleteKnowledgeBase(deletingKb.value.id).then((res: any) => {
if (res.success) {
MessagePlugin.success('已删除')
deleteVisible.value = false
deletingKb.value = null
fetchList()
} else {
MessagePlugin.error(res.message || '删除失败')
}
}).catch((e: any) => MessagePlugin.error(e?.message || '删除失败'))
}).catch((e: any) => {
MessagePlugin.error(e?.message || '删除失败')
})
}
const isInitialized = (kb: KB) => {
@@ -173,39 +222,77 @@ const hasUninitializedKbs = computed(() => {
return kbs.value.some(kb => !isInitialized(kb))
})
const handleCardClick = (kb: KB) => {
if (isInitialized(kb)) {
goDetail(kb.id)
} else {
goSettings(kb.id)
}
}
const goDetail = (id: string) => {
router.push(`/platform/knowledge-bases/${id}`)
}
const goSettings = (id: string) => {
// 使用模态框打开设置,而不是路由导航
// 使用模态框打开设置
uiStore.openKBSettings(id)
}
// 监听知识库设置模态框关闭,刷新列表
watch(() => uiStore.showKBSettingsModal, (newVal, oldVal) => {
// 当模态框从打开变为关闭时,刷新列表
if (oldVal && !newVal) {
fetchList()
}
})
// 知识库编辑器成功回调(创建或编辑成功)
const handleKBEditorSuccess = (kbId: string) => {
console.log('知识库操作成功:', kbId)
fetchList()
}
</script>
<style scoped lang="less">
.kb-list-container {
padding: 20px;
background: #fff;
margin: 0 20px 0 20px;
padding: 24px 44px;
// background: #fff;
margin: 0 20px;
height: calc(100vh);
overflow-y: auto;
box-sizing: border-box;
flex: 1;
}
.header {
display: flex;
align-items: center;
justify-content: space-between;
margin-bottom: 16px;
h2 { margin: 0; font-size: 20px; font-weight: 600; }
margin-bottom: 20px;
h2 {
margin: 0;
color: #000000e6;
font-family: "PingFang SC";
font-size: 24px;
font-weight: 600;
line-height: 32px;
}
}
.create-btn {
padding: 8px 20px;
height: 36px;
background: #07c05f;
color: #fff;
border: none;
border-radius: 6px;
font-family: "PingFang SC";
font-size: 14px;
font-weight: 500;
cursor: pointer;
transition: all 0.2s ease;
&:hover {
background: #05a04f;
}
&:active {
background: #048f45;
}
}
.warning-banner {
@@ -213,11 +300,12 @@ watch(() => uiStore.showKBSettingsModal, (newVal, oldVal) => {
align-items: center;
gap: 8px;
padding: 12px 16px;
margin-bottom: 16px;
margin-bottom: 20px;
background: #fff7e6;
border: 1px solid #ffd591;
border-radius: 6px;
color: #d46b08;
font-family: "PingFang SC";
font-size: 14px;
.t-icon {
@@ -226,28 +314,349 @@ watch(() => uiStore.showKBSettingsModal, (newVal, oldVal) => {
}
}
.status-cell {
.kb-card-wrap {
display: grid;
gap: 20px;
grid-template-columns: 1fr;
}
.kb-card {
border: 2px solid #fbfbfb;
border-radius: 6px;
overflow: hidden;
box-sizing: border-box;
box-shadow: 0 0 8px 0 #00000005;
background: #fff;
position: relative;
cursor: pointer;
transition: all 0.2s ease;
padding: 12px 16px 14px;
display: flex;
align-items: center;
gap: 8px;
.warning-icon {
color: #ff8800;
cursor: pointer;
font-size: 16px;
font-weight: bold;
transition: color 0.2s;
&:hover {
color: #d46b08;
}
flex-direction: column;
min-height: 150px;
&:hover {
border-color: #07c05f;
}
&.uninitialized {
opacity: 0.9;
}
}
.description-cell {
.description-text {
color: #000000e6;
.card-header {
display: flex;
justify-content: space-between;
align-items: center;
margin-bottom: 8px;
}
.card-title {
color: #000000e6;
font-family: "PingFang SC";
font-size: 14px;
font-weight: 600;
line-height: 22px;
overflow: hidden;
text-overflow: ellipsis;
white-space: nowrap;
flex: 1;
margin-right: 8px;
}
.more-wrap {
display: flex;
width: 28px;
height: 28px;
justify-content: center;
align-items: center;
border-radius: 3px;
cursor: pointer;
flex-shrink: 0;
&:hover {
background: #e7e7e7;
}
&.active-more {
background: #e7e7e7;
}
.more-icon {
width: 14px;
height: 14px;
}
}
.card-content {
flex: 1;
margin-bottom: 10px;
}
.card-description {
display: -webkit-box;
-webkit-box-orient: vertical;
-webkit-line-clamp: 2;
overflow: hidden;
color: #00000066;
font-family: "PingFang SC";
font-size: 12px;
font-weight: 400;
line-height: 20px;
min-height: 40px;
}
.card-bottom {
display: flex;
align-items: center;
justify-content: space-between;
padding: 8px 0 0;
border-top: 1px solid #f0f0f0;
margin-top: auto;
}
.status-badge {
display: flex;
align-items: center;
gap: 3px;
padding: 2px 8px;
border-radius: 3px;
font-family: "PingFang SC";
font-size: 12px;
font-weight: 500;
&.initialized {
background: #07c05f1a;
color: #07c05f;
}
&.uninitialized {
background: #fff7e6;
color: #d46b08;
}
.warning-icon {
font-size: 12px;
}
}
.card-time {
color: #00000066;
font-family: "PingFang SC";
font-size: 12px;
font-weight: 400;
}
.empty-state {
flex: 1;
display: flex;
flex-direction: column;
justify-content: center;
align-items: center;
padding: 60px 20px;
.empty-img {
width: 162px;
height: 162px;
margin-bottom: 20px;
}
.empty-txt {
color: #00000099;
font-family: "PingFang SC";
font-size: 16px;
font-weight: 600;
line-height: 26px;
margin-bottom: 8px;
}
.empty-desc {
color: #00000066;
font-family: "PingFang SC";
font-size: 14px;
font-weight: 400;
line-height: 22px;
}
}
// 响应式布局
@media (min-width: 900px) {
.kb-card-wrap {
grid-template-columns: repeat(2, 1fr);
}
}
@media (min-width: 1250px) {
.kb-card-wrap {
grid-template-columns: repeat(3, 1fr);
}
}
@media (min-width: 1600px) {
.kb-card-wrap {
grid-template-columns: repeat(4, 1fr);
}
}
// 删除确认对话框样式
:deep(.del-knowledge-dialog) {
padding: 0px !important;
border-radius: 6px !important;
.t-dialog__header {
display: none;
}
.t-dialog__body {
padding: 16px;
}
.t-dialog__footer {
padding: 0;
}
}
:deep(.t-dialog__position.t-dialog--top) {
padding-top: 40vh !important;
}
.circle-wrap {
.dialog-header {
display: flex;
align-items: center;
margin-bottom: 8px;
}
.circle-img {
width: 20px;
height: 20px;
margin-right: 8px;
}
.circle-title {
color: #000000e6;
font-family: "PingFang SC";
font-size: 16px;
font-weight: 600;
line-height: 24px;
}
.del-circle-txt {
color: #00000099;
font-family: "PingFang SC";
font-size: 14px;
font-weight: 400;
line-height: 22px;
display: inline-block;
margin-left: 29px;
margin-bottom: 21px;
}
.circle-btn {
height: 22px;
width: 100%;
display: flex;
justify-content: flex-end;
}
.circle-btn-txt {
color: #000000e6;
font-family: "PingFang SC";
font-size: 14px;
font-weight: 400;
line-height: 22px;
cursor: pointer;
&:hover {
opacity: 0.8;
}
}
.confirm {
color: #FA5151;
margin-left: 40px;
&:hover {
opacity: 0.8;
}
}
}
</style>
<style lang="less">
// 更多操作弹窗样式
.card-more-popup {
z-index: 99 !important;
.t-popup__content {
padding: 4px 0 !important;
margin-top: 4px !important;
min-width: 120px;
}
}
.popup-menu {
display: flex;
flex-direction: column;
}
.popup-menu-item {
display: flex;
align-items: center;
padding: 8px 16px;
cursor: pointer;
transition: all 0.2s ease;
color: #000000e6;
font-family: "PingFang SC";
font-size: 14px;
font-weight: 400;
.menu-icon {
margin-right: 8px;
font-size: 16px;
}
&:hover {
background: #f5f5f5;
}
&.delete {
color: #000000e6;
&:hover {
background: #fff1f0;
color: #FA5151;
.menu-icon {
color: #FA5151;
}
}
}
}
// 创建对话框样式优化
.create-kb-dialog {
.t-form-item__label {
font-family: "PingFang SC";
font-size: 14px;
font-weight: 500;
color: #000000e6;
}
.t-input,
.t-textarea {
font-family: "PingFang SC";
}
.t-button--theme-primary {
background-color: #07c05f;
border-color: #07c05f;
&:hover {
background-color: #05a04f;
border-color: #05a04f;
}
}
}
</style>
@@ -1,668 +0,0 @@
<template>
<Teleport to="body">
<Transition name="modal">
<div v-if="visible" class="settings-overlay" @click.self="handleClose">
<div class="settings-modal">
<!-- 关闭按钮 -->
<button class="close-btn" @click="handleClose" aria-label="关闭设置">
<svg width="20" height="20" viewBox="0 0 20 20" fill="currentColor">
<path d="M15 5L5 15M5 5L15 15" stroke="currentColor" stroke-width="2" stroke-linecap="round"/>
</svg>
</button>
<div class="settings-container">
<!-- 左侧导航 -->
<div class="settings-sidebar">
<div class="sidebar-header">
<h2 class="sidebar-title">知识库设置</h2>
</div>
<div class="settings-nav">
<div
v-for="(item, index) in navItems"
:key="index"
:class="['nav-item', { 'active': currentSection === item.key }]"
@click="currentSection = item.key"
>
<span class="nav-icon">{{ item.icon }}</span>
<span class="nav-label">{{ item.label }}</span>
</div>
</div>
</div>
<!-- 右侧内容区域 -->
<div class="settings-content">
<div class="content-wrapper">
<!-- 基本信息 -->
<div v-show="currentSection === 'basic'" class="section">
<KBBasicInfo
v-if="kbData"
:kb-id="kbId"
:name="kbData.name"
:description="kbData.description"
@update:name="kbData.name = $event"
@update:description="kbData.description = $event"
/>
</div>
<!-- 模型配置 -->
<div v-show="currentSection === 'models'" class="section">
<KBModelConfig
ref="modelConfigRef"
v-if="configData"
:config="modelConfig"
:has-files="configData.hasFiles"
:all-models="allModels"
@update:config="handleModelConfigUpdate"
/>
</div>
<!-- 分块设置 -->
<div v-show="currentSection === 'chunking'" class="section">
<KBChunkingSettings
v-if="configData"
:config="chunkingConfig"
@update:config="handleChunkingConfigUpdate"
/>
</div>
<!-- 高级设置 -->
<div v-show="currentSection === 'advanced'" class="section">
<KBAdvancedSettings
ref="advancedSettingsRef"
v-if="configData"
:multimodal="multimodalConfig"
:node-extract="nodeExtractConfig"
:all-models="allModels"
@update:multimodal="handleMultimodalUpdate"
@update:nodeExtract="handleNodeExtractUpdate"
/>
</div>
</div>
<!-- 保存按钮 -->
<div class="settings-footer">
<t-button theme="default" variant="outline" @click="handleClose">
取消
</t-button>
<t-button theme="primary" @click="handleSave" :loading="saving">
保存配置
</t-button>
</div>
</div>
</div>
</div>
</div>
</Transition>
</Teleport>
</template>
<script setup lang="ts">
import { ref, computed, watch, onMounted } from 'vue'
import { useUIStore } from '@/stores/ui'
import { MessagePlugin } from 'tdesign-vue-next'
import { getKnowledgeBaseById, listKnowledgeFiles } from '@/api/knowledge-base'
import { updateKBConfig, type KBModelConfigRequest } from '@/api/initialization'
import { listModels } from '@/api/model'
import KBBasicInfo from './settings/KBBasicInfo.vue'
import KBModelConfig from './settings/KBModelConfig.vue'
import KBChunkingSettings from './settings/KBChunkingSettings.vue'
import KBAdvancedSettings from './settings/KBAdvancedSettings.vue'
const uiStore = useUIStore()
const visible = computed(() => uiStore.showKBSettingsModal)
const kbId = computed(() => uiStore.currentKBId || '')
const currentSection = ref<string>('basic')
const saving = ref(false)
const loading = ref(false)
const allModels = ref<any[]>([])
const navItems = [
{ key: 'basic', icon: 'ℹ️', label: '基本信息' },
{ key: 'models', icon: '🤖', label: '模型配置' },
{ key: 'chunking', icon: '📄', label: '分块设置' },
{ key: 'advanced', icon: '⚙️', label: '高级设置' }
]
// 知识库基本数据
const kbData = ref<{
name: string
description?: string
} | null>(null)
// 配置数据(简化为只需要 hasFiles 标记)
const configData = ref<{ hasFiles: boolean } | null>(null)
// 模型配置引用
const modelConfigRef = ref<InstanceType<typeof KBModelConfig>>()
const advancedSettingsRef = ref<InstanceType<typeof KBAdvancedSettings>>()
// 模型配置
const modelConfig = ref({
llmModelId: '',
embeddingModelId: '',
rerankModelId: ''
})
// 分块配置
const chunkingConfig = ref({
chunkSize: 512,
chunkOverlap: 100,
separators: ['\n\n', '\n', '。', '!', '?', ';', ';']
})
// 多模态配置
const multimodalConfig = ref({
enabled: false,
storageType: 'minio' as 'minio' | 'cos',
vllmModelId: '',
minio: {
bucketName: '',
useSSL: false,
pathPrefix: ''
},
cos: {
secretId: '',
secretKey: '',
region: '',
bucketName: '',
appId: '',
pathPrefix: ''
}
})
// 知识图谱配置
const nodeExtractConfig = ref({
enabled: false,
text: '',
tags: [] as string[]
})
// 加载所有模型(统一加载,避免重复调用)
const loadAllModels = async () => {
try {
const models = await listModels()
allModels.value = models || []
return models || []
} catch (error) {
console.error('加载模型列表失败:', error)
MessagePlugin.error('加载模型列表失败')
allModels.value = []
return []
}
}
// 加载知识库数据
const loadKBData = async () => {
if (!kbId.value) return
loading.value = true
try {
// 并行加载知识库信息和所有模型(只需要一个接口!)
const [kbInfo, models] = await Promise.all([
getKnowledgeBaseById(kbId.value),
loadAllModels()
])
if (!kbInfo || !kbInfo.data) {
throw new Error('知识库不存在')
}
const kb = kbInfo.data
// 设置基本信息
kbData.value = {
name: kb.name || '',
description: kb.description || ''
}
// 设置模型ID(直接从知识库数据中获取)
modelConfig.value = {
llmModelId: kb.summary_model_id || '',
embeddingModelId: kb.embedding_model_id || '',
rerankModelId: kb.rerank_model_id || ''
}
// 设置分块配置
chunkingConfig.value = {
chunkSize: kb.chunking_config?.chunk_size || 512,
chunkOverlap: kb.chunking_config?.chunk_overlap || 100,
separators: kb.chunking_config?.separators || ['\n\n', '\n', '。', '!', '?', ';', ';']
}
// 设置多模态配置
// 判断是否真的有存储配置:provider 不为空且有实际的存储信息
const hasStorage = kb.cos_config?.provider && (
(kb.cos_config.provider === 'minio' && kb.cos_config.bucket_name) ||
(kb.cos_config.provider === 'cos' && kb.cos_config.bucket_name)
)
const hasVLM = !!kb.vlm_model_id
multimodalConfig.value = {
enabled: !!(hasStorage || hasVLM),
storageType: (kb.cos_config?.provider || 'minio') as 'minio' | 'cos',
vllmModelId: kb.vlm_model_id || '',
minio: {
bucketName: kb.cos_config?.bucket_name || '',
useSSL: false,
pathPrefix: kb.cos_config?.path_prefix || ''
},
cos: {
secretId: kb.cos_config?.secret_id || '',
secretKey: kb.cos_config?.secret_key || '',
region: kb.cos_config?.region || '',
bucketName: kb.cos_config?.bucket_name || '',
appId: kb.cos_config?.app_id || '',
pathPrefix: kb.cos_config?.path_prefix || ''
}
}
// 设置知识图谱配置
if (kb.extract_config) {
nodeExtractConfig.value = {
enabled: true,
text: kb.extract_config.text || '',
tags: kb.extract_config.tags || []
}
} else {
nodeExtractConfig.value = {
enabled: false,
text: '',
tags: []
}
}
// 检查是否有文件(用于判断是否可以修改Embedding模型)
const knowledgeList = await listKnowledgeFiles(kbId.value, { page: 1, page_size: 1 })
const hasFiles = knowledgeList && knowledgeList.data && knowledgeList.data.items && knowledgeList.data.items.length > 0
// 设置 configData(用于传递 hasFiles 状态)
configData.value = { hasFiles } as any
} catch (error) {
console.error('加载知识库数据失败:', error)
MessagePlugin.error('加载知识库数据失败')
} finally {
loading.value = false
}
}
// 不再需要 mapModelIds,直接使用知识库返回的模型ID
// 处理模型配置更新
const handleModelConfigUpdate = (newConfig: typeof modelConfig.value) => {
modelConfig.value = newConfig
}
// 处理分块配置更新
const handleChunkingConfigUpdate = (newConfig: typeof chunkingConfig.value) => {
chunkingConfig.value = newConfig
}
// 处理多模态配置更新
const handleMultimodalUpdate = (newConfig: typeof multimodalConfig.value) => {
multimodalConfig.value = newConfig
}
// 处理知识图谱配置更新
const handleNodeExtractUpdate = (newConfig: typeof nodeExtractConfig.value) => {
nodeExtractConfig.value = newConfig
}
// 保存配置
const handleSave = async () => {
if (saving.value || !kbId.value) return
// 验证必填项 - 模型配置
if (!modelConfig.value.llmModelId) {
MessagePlugin.warning('请选择LLM模型')
currentSection.value = 'models'
return
}
if (!modelConfig.value.embeddingModelId) {
MessagePlugin.warning('请选择Embedding模型')
currentSection.value = 'models'
return
}
// 验证多模态配置
if (multimodalConfig.value.enabled) {
// 验证VLLM模型(必选)
if (!multimodalConfig.value.vllmModelId) {
MessagePlugin.warning('多模态功能已启用,请选择VLLM视觉模型')
currentSection.value = 'advanced'
return
}
// 验证存储配置
if (multimodalConfig.value.storageType === 'minio') {
if (!multimodalConfig.value.minio.bucketName) {
MessagePlugin.warning('使用MinIO存储时,Bucket名称为必填项')
currentSection.value = 'advanced'
return
}
} else if (multimodalConfig.value.storageType === 'cos') {
const cos = multimodalConfig.value.cos
if (!cos.secretId) {
MessagePlugin.warning('使用腾讯云COS时,SecretId为必填项')
currentSection.value = 'advanced'
return
}
if (!cos.secretKey) {
MessagePlugin.warning('使用腾讯云COS时,SecretKey为必填项')
currentSection.value = 'advanced'
return
}
if (!cos.region) {
MessagePlugin.warning('使用腾讯云COS时,地域为必填项')
currentSection.value = 'advanced'
return
}
if (!cos.bucketName) {
MessagePlugin.warning('使用腾讯云COS时,Bucket名称为必填项')
currentSection.value = 'advanced'
return
}
if (!cos.appId) {
MessagePlugin.warning('使用腾讯云COS时,AppId为必填项')
currentSection.value = 'advanced'
return
}
}
}
saving.value = true
try {
// 构建配置对象(简化版,只传模型ID)
const config: KBModelConfigRequest = {
llmModelId: modelConfig.value.llmModelId,
embeddingModelId: modelConfig.value.embeddingModelId,
rerankModelId: modelConfig.value.rerankModelId || '',
vllmModelId: multimodalConfig.value.vllmModelId || '',
documentSplitting: {
chunkSize: chunkingConfig.value.chunkSize,
chunkOverlap: chunkingConfig.value.chunkOverlap,
separators: chunkingConfig.value.separators
},
multimodal: {
enabled: multimodalConfig.value.enabled,
storageType: multimodalConfig.value.enabled ? multimodalConfig.value.storageType : 'minio',
cos: multimodalConfig.value.enabled && multimodalConfig.value.storageType === 'cos' ? multimodalConfig.value.cos : undefined,
minio: multimodalConfig.value.enabled && multimodalConfig.value.storageType === 'minio' ? multimodalConfig.value.minio : undefined
},
nodeExtract: {
enabled: nodeExtractConfig.value.enabled,
text: nodeExtractConfig.value.text,
tags: nodeExtractConfig.value.tags,
nodes: [],
relations: []
}
}
// 保存配置
await updateKBConfig(kbId.value, config)
MessagePlugin.success('配置保存成功')
handleClose()
} catch (error) {
console.error('保存配置失败:', error)
MessagePlugin.error('保存配置失败')
} finally {
saving.value = false
}
}
// 处理关闭
const handleClose = () => {
uiStore.closeKBSettings()
// 重置当前选中的section
setTimeout(() => {
currentSection.value = 'basic'
}, 300)
}
// 监听visible变化,当打开时加载数据
watch(visible, (newVal) => {
if (newVal && kbId.value) {
loadKBData()
}
})
// 监听全局设置关闭事件,刷新模型列表
watch(() => uiStore.showSettingsModal, (newVal, oldVal) => {
// 当全局设置从打开变为关闭,且知识库设置仍然打开时,刷新模型列表
if (oldVal && !newVal && visible.value) {
// 统一刷新所有模型列表(只调用一次API)
loadAllModels()
}
})
onMounted(() => {
if (visible.value && kbId.value) {
loadKBData()
}
})
</script>
<style lang="less" scoped>
/* 遮罩层 */
.settings-overlay {
position: fixed;
inset: 0;
z-index: 999;
background: rgba(0, 0, 0, 0.4);
display: flex;
align-items: center;
justify-content: center;
padding: 20px;
}
/* 弹窗容器 */
.settings-modal {
position: relative;
width: 100%;
max-width: 900px;
height: 700px;
background: #ffffff;
border-radius: 12px;
box-shadow: 0 6px 28px rgba(15, 23, 42, 0.08);
overflow: hidden;
display: flex;
flex-direction: column;
}
/* 关闭按钮 */
.close-btn {
position: absolute;
top: 16px;
right: 16px;
width: 32px;
height: 32px;
border: none;
background: transparent;
color: #666666;
cursor: pointer;
border-radius: 6px;
display: flex;
align-items: center;
justify-content: center;
transition: all 0.2s ease;
z-index: 10;
&:hover {
background: #f5f5f5;
color: #333333;
}
}
.settings-container {
display: flex;
height: 100%;
width: 100%;
overflow: hidden;
}
/* 左侧导航栏 */
.settings-sidebar {
width: 220px;
background-color: #f8f9fa;
border-right: 1px solid #e5e7eb;
flex-shrink: 0;
overflow-y: auto;
display: flex;
flex-direction: column;
}
.sidebar-header {
padding: 24px 16px 16px;
border-bottom: 1px solid #e5e7eb;
}
.sidebar-title {
font-size: 18px;
font-weight: 600;
color: #333333;
margin: 0;
}
.settings-nav {
padding: 16px 8px;
flex: 1;
}
.nav-item {
display: flex;
align-items: center;
padding: 10px 16px;
margin-bottom: 4px;
border-radius: 6px;
cursor: pointer;
color: #666666;
font-size: 14px;
transition: all 0.2s ease;
user-select: none;
&:hover {
background-color: #e8f5ed;
color: #333333;
}
&.active {
background-color: rgba(7, 192, 95, 0.1);
color: #07C05F;
font-weight: 500;
}
}
.nav-icon {
margin-right: 12px;
font-size: 16px;
display: flex;
align-items: center;
justify-content: center;
width: 20px;
height: 20px;
}
.nav-label {
flex: 1;
}
/* 右侧内容区域 */
.settings-content {
flex: 1;
overflow-y: auto;
background-color: #ffffff;
display: flex;
flex-direction: column;
}
.content-wrapper {
flex: 1;
max-width: 600px;
padding: 40px 48px;
overflow-y: auto;
}
.section {
animation: fadeIn 0.3s ease;
}
@keyframes fadeIn {
from {
opacity: 0;
transform: translateY(10px);
}
to {
opacity: 1;
transform: translateY(0);
}
}
/* 底部保存按钮 */
.settings-footer {
padding: 16px 48px;
border-top: 1px solid #e5e7eb;
display: flex;
justify-content: flex-end;
gap: 12px;
background: #ffffff;
}
/* 弹窗动画 */
.modal-enter-active,
.modal-leave-active {
transition: opacity 0.2s ease;
}
.modal-enter-active .settings-modal,
.modal-leave-active .settings-modal {
transition: transform 0.2s ease, opacity 0.2s ease;
}
.modal-enter-from,
.modal-leave-to {
opacity: 0;
}
.modal-enter-from .settings-modal,
.modal-leave-to .settings-modal {
transform: scale(0.95);
opacity: 0;
}
/* 滚动条样式 */
.settings-sidebar::-webkit-scrollbar,
.settings-content::-webkit-scrollbar,
.content-wrapper::-webkit-scrollbar {
width: 6px;
}
.settings-sidebar::-webkit-scrollbar-track {
background: #f8f9fa;
}
.settings-sidebar::-webkit-scrollbar-thumb {
background: #d0d0d0;
border-radius: 3px;
}
.settings-sidebar::-webkit-scrollbar-thumb:hover {
background: #b0b0b0;
}
.content-wrapper::-webkit-scrollbar-track {
background: #ffffff;
}
.content-wrapper::-webkit-scrollbar-thumb {
background: #d0d0d0;
border-radius: 3px;
}
.content-wrapper::-webkit-scrollbar-thumb:hover {
background: #b0b0b0;
}
</style>
@@ -1,198 +0,0 @@
<template>
<div class="kb-basic-info">
<div class="section-header">
<h2>基本信息</h2>
<p class="section-description">配置知识库的基本信息</p>
</div>
<div class="settings-group">
<!-- 知识库名称 -->
<div class="setting-row">
<div class="setting-info">
<label>知识库名称 <span class="required">*</span></label>
<p class="desc">知识库的显示名称(最多50个字符)</p>
</div>
<div class="setting-control">
<t-input
v-model="localName"
placeholder="请输入知识库名称"
maxlength="50"
show-word-limit
@blur="handleNameChange"
style="width: 280px;"
/>
</div>
</div>
<!-- 知识库描述 -->
<div class="setting-row">
<div class="setting-info">
<label>知识库描述</label>
<p class="desc">知识库的详细描述信息(最多200个字符)</p>
</div>
<div class="setting-control">
<t-textarea
v-model="localDescription"
placeholder="请输入知识库描述"
maxlength="200"
show-word-limit
:autosize="{ minRows: 3, maxRows: 6 }"
@blur="handleDescriptionChange"
style="width: 280px;"
/>
</div>
</div>
</div>
</div>
</template>
<script setup lang="ts">
import { ref, watch } from 'vue'
import { MessagePlugin } from 'tdesign-vue-next'
import { updateKnowledgeBase } from '@/api/knowledge-base'
interface Props {
kbId: string
name: string
description?: string
}
const props = defineProps<Props>()
const emit = defineEmits<{
'update:name': [value: string]
'update:description': [value: string]
}>()
const localName = ref(props.name)
const localDescription = ref(props.description || '')
const saving = ref(false)
// 监听props变化
watch(() => props.name, (newVal) => {
localName.value = newVal
})
watch(() => props.description, (newVal) => {
localDescription.value = newVal || ''
})
// 保存基本信息
const saveBasicInfo = async () => {
if (saving.value) return
if (!localName.value.trim()) {
MessagePlugin.warning('知识库名称不能为空')
return
}
saving.value = true
try {
await updateKnowledgeBase(props.kbId, {
name: localName.value,
description: localDescription.value,
config: {} // 这里只更新基本信息,不更新配置
})
emit('update:name', localName.value)
emit('update:description', localDescription.value)
MessagePlugin.success('保存成功')
} catch (error) {
console.error('保存知识库基本信息失败:', error)
MessagePlugin.error('保存失败')
} finally {
saving.value = false
}
}
// 处理名称变化
const handleNameChange = () => {
if (localName.value !== props.name) {
saveBasicInfo()
}
}
// 处理描述变化
const handleDescriptionChange = () => {
if (localDescription.value !== props.description) {
saveBasicInfo()
}
}
</script>
<style lang="less" scoped>
.kb-basic-info {
width: 100%;
}
.section-header {
margin-bottom: 32px;
h2 {
font-size: 20px;
font-weight: 600;
color: #333333;
margin: 0 0 8px 0;
}
.section-description {
font-size: 14px;
color: #666666;
margin: 0;
line-height: 1.5;
}
}
.settings-group {
display: flex;
flex-direction: column;
gap: 0;
}
.setting-row {
display: flex;
align-items: flex-start;
justify-content: space-between;
padding: 20px 0;
border-bottom: 1px solid #e5e7eb;
&:last-child {
border-bottom: none;
}
}
.setting-info {
flex: 1;
max-width: 65%;
padding-right: 24px;
label {
font-size: 15px;
font-weight: 500;
color: #333333;
display: block;
margin-bottom: 4px;
.required {
color: #fa5151;
margin-left: 2px;
}
}
.desc {
font-size: 13px;
color: #666666;
margin: 0;
line-height: 1.5;
}
}
.setting-control {
flex-shrink: 0;
min-width: 280px;
display: flex;
justify-content: flex-end;
align-items: center;
}
</style>
+2 -1
View File
@@ -243,11 +243,12 @@ onUnmounted(() => {
position: fixed;
inset: 0;
z-index: 1000;
background: transparent;
background: rgba(0, 0, 0, 0.5);
display: flex;
align-items: center;
justify-content: center;
padding: 20px;
backdrop-filter: blur(4px);
}
/* 弹窗容器 */
+38 -9
View File
@@ -28,7 +28,9 @@ type AgentEngine struct {
chatModel chat.Chat
knowledgeService interfaces.KnowledgeBaseService
eventBus *event.EventBus
knowledgeBasesInfo []*KnowledgeBaseInfo // Detailed knowledge base information for prompt
knowledgeBasesInfo []*KnowledgeBaseInfo // Detailed knowledge base information for prompt
contextManager interfaces.ContextManager // Context manager for writing agent conversation to LLM context
sessionID string // Session ID for context management
}
// listToolNames returns tool.function names for logging
@@ -48,6 +50,8 @@ func NewAgentEngine(
knowledgeService interfaces.KnowledgeBaseService,
eventBus *event.EventBus,
knowledgeBasesInfo []*KnowledgeBaseInfo,
contextManager interfaces.ContextManager,
sessionID string,
) *AgentEngine {
if eventBus == nil {
eventBus = event.NewEventBus()
@@ -59,14 +63,16 @@ func NewAgentEngine(
knowledgeService: knowledgeService,
eventBus: eventBus,
knowledgeBasesInfo: knowledgeBasesInfo,
contextManager: contextManager,
sessionID: sessionID,
}
}
// Execute executes the agent with conversation history and streaming output
// All events are emitted to EventBus and handled by subscribers (like Handler layer)
func (e *AgentEngine) Execute(ctx context.Context, sessionID, query string, llmContext []chat.Message) (*types.AgentState, error) {
func (e *AgentEngine) Execute(ctx context.Context, sessionID, messageID, query string, llmContext []chat.Message) (*types.AgentState, error) {
logger.Infof(ctx, "========== Agent Execution Started ==========")
logger.Infof(ctx, "[Agent] SessionID: %s", sessionID)
logger.Infof(ctx, "[Agent] SessionID: %s, MessageID: %s", sessionID, messageID)
logger.Infof(ctx, "[Agent] User Query: %s", query)
logger.Infof(ctx, "[Agent] LLM Context Messages: %d", len(llmContext))
@@ -92,7 +98,7 @@ func (e *AgentEngine) Execute(ctx context.Context, sessionID, query string, llmC
tools := e.buildToolsForLLM()
logger.Infof(ctx, "[Agent] Tools enabled (%d): %s", len(tools), strings.Join(listToolNames(tools), ", "))
_, err := e.executeLoop(ctx, state, query, messages, tools, sessionID)
_, err := e.executeLoop(ctx, state, query, messages, tools, sessionID, messageID)
if err != nil {
logger.Errorf(ctx, "[Agent] Execution failed: %v", err)
e.eventBus.Emit(ctx, event.Event{
@@ -123,6 +129,7 @@ func (e *AgentEngine) executeLoop(
messages []chat.Message,
tools []chat.Tool,
sessionID string,
messageID string,
) (*types.AgentState, error) {
startTime := time.Now()
for state.CurrentRound < e.config.MaxIterations {
@@ -336,8 +343,8 @@ func (e *AgentEngine) executeLoop(
}
state.RoundSteps = append(state.RoundSteps, step)
// 4. Observe: Add tool results to messages
messages = e.appendToolResults(messages, step)
// 4. Observe: Add tool results to messages and write to context
messages = e.appendToolResults(ctx, messages, step)
// 5. Check if we should continue
state.CurrentRound++
}
@@ -371,6 +378,7 @@ func (e *AgentEngine) executeLoop(
AgentSteps: state.RoundSteps, // Include detailed execution steps for message storage
TotalSteps: len(state.RoundSteps),
TotalDuration: time.Since(startTime).Milliseconds(),
MessageID: messageID, // Include message ID for proper message update
},
})
@@ -397,7 +405,8 @@ func (e *AgentEngine) buildToolsForLLM() []chat.Tool {
}
// appendToolResults adds tool results to the message history following OpenAI's tool calling format
func (e *AgentEngine) appendToolResults(messages []chat.Message, step types.AgentStep) []chat.Message {
// Also writes these messages to the context manager for persistence
func (e *AgentEngine) appendToolResults(ctx context.Context, messages []chat.Message, step types.AgentStep) []chat.Message {
// Add assistant message with tool calls (if any)
if step.Thought != "" || len(step.ToolCalls) > 0 {
assistantMsg := chat.Message{
@@ -424,6 +433,15 @@ func (e *AgentEngine) appendToolResults(messages []chat.Message, step types.Agen
}
messages = append(messages, assistantMsg)
// Write assistant message to context
if e.contextManager != nil {
if err := e.contextManager.AddMessage(ctx, e.sessionID, assistantMsg); err != nil {
logger.Warnf(ctx, "[Agent] Failed to add assistant message to context: %v", err)
} else {
logger.Debugf(ctx, "[Agent] Added assistant message to context (session: %s)", e.sessionID)
}
}
}
// Add tool result messages (role: "tool", following OpenAI format)
@@ -433,12 +451,23 @@ func (e *AgentEngine) appendToolResults(messages []chat.Message, step types.Agen
resultContent = fmt.Sprintf("Error: %s", toolCall.Result.Error)
}
messages = append(messages, chat.Message{
toolMsg := chat.Message{
Role: "tool",
Content: resultContent,
ToolCallID: toolCall.ID,
Name: toolCall.Name,
})
}
messages = append(messages, toolMsg)
// Write tool message to context
if e.contextManager != nil {
if err := e.contextManager.AddMessage(ctx, e.sessionID, toolMsg); err != nil {
logger.Warnf(ctx, "[Agent] Failed to add tool message to context: %v", err)
} else {
logger.Debugf(ctx, "[Agent] Added tool message to context (session: %s, tool: %s)", e.sessionID, toolCall.Name)
}
}
}
return messages
@@ -53,6 +53,8 @@ func (s *agentService) CreateAgentEngine(
ctx context.Context,
config *types.AgentConfig,
eventBus *event.EventBus,
contextManager interfaces.ContextManager,
sessionID string,
) (interfaces.AgentEngine, error) {
logger.Infof(ctx, "Creating agent engine with custom EventBus")
@@ -103,7 +105,7 @@ func (s *agentService) CreateAgentEngine(
}
}
// Create engine with provided EventBus
// Create engine with provided EventBus and contextManager
engine := agent.NewAgentEngine(
config,
chatModel,
@@ -111,6 +113,8 @@ func (s *agentService) CreateAgentEngine(
s.knowledgeBaseService,
eventBus,
kbInfos,
contextManager,
sessionID,
)
return engine, nil
@@ -3,6 +3,7 @@ package chatpipline
import (
"context"
"strings"
"sync"
"github.com/Tencent/WeKnora/internal/config"
"github.com/Tencent/WeKnora/internal/logger"
@@ -40,6 +41,21 @@ func (p *PluginSearch) ActivationEvents() []types.EventType {
func (p *PluginSearch) OnEvent(ctx context.Context,
eventType types.EventType, chatManage *types.ChatManage, next func() *PluginError,
) *PluginError {
// Get knowledge base IDs list
knowledgeBaseIDs := chatManage.KnowledgeBaseIDs
if len(knowledgeBaseIDs) == 0 && chatManage.KnowledgeBaseID != "" {
// Fall back to single knowledge base
knowledgeBaseIDs = []string{chatManage.KnowledgeBaseID}
logger.Infof(ctx, "No KnowledgeBaseIDs provided, falling back to single KB: %s", chatManage.KnowledgeBaseID)
}
if len(knowledgeBaseIDs) == 0 {
logger.Errorf(ctx, "No knowledge base IDs available for search")
return ErrSearch.WithError(nil)
}
logger.Infof(ctx, "Searching across %d knowledge base(s): %v", len(knowledgeBaseIDs), knowledgeBaseIDs)
// Prepare search parameters
searchParams := types.SearchParams{
QueryText: strings.TrimSpace(chatManage.RewriteQuery),
@@ -49,14 +65,34 @@ func (p *PluginSearch) OnEvent(ctx context.Context,
}
logger.Infof(ctx, "Search parameters: %v", searchParams)
// Perform initial hybrid search
searchResults, err := p.knowledgeBaseService.HybridSearch(ctx, chatManage.KnowledgeBaseID, searchParams)
logger.Infof(ctx, "Search results count: %d, error: %v", len(searchResults), err)
if err != nil {
return ErrSearch.WithError(err)
// Parallel search across multiple knowledge bases
var wg sync.WaitGroup
var mu sync.Mutex
var allResults []*types.SearchResult
for _, kbID := range knowledgeBaseIDs {
wg.Add(1)
go func(knowledgeBaseID string) {
defer wg.Done()
results, err := p.knowledgeBaseService.HybridSearch(ctx, knowledgeBaseID, searchParams)
if err != nil {
logger.Errorf(ctx, "Failed to search KB %s: %v", knowledgeBaseID, err)
return
}
logger.Infof(ctx, "KB %s search results count: %d", knowledgeBaseID, len(results))
mu.Lock()
allResults = append(allResults, results...)
mu.Unlock()
}(kbID)
}
chatManage.SearchResult = searchResults
logger.Infof(ctx, "Search result count: %d", len(chatManage.SearchResult))
wg.Wait()
logger.Infof(ctx, "Total search results from all KBs: %d", len(allResults))
chatManage.SearchResult = allResults
// Add relevant results from chat history
historyResult := p.getSearchResultFromHistory(chatManage)
@@ -68,17 +104,38 @@ func (p *PluginSearch) OnEvent(ctx context.Context,
// Try search with processed query if different from rewrite query
if chatManage.RewriteQuery != chatManage.ProcessedQuery {
searchParams.QueryText = strings.TrimSpace(chatManage.ProcessedQuery)
searchResults, err = p.knowledgeBaseService.HybridSearch(ctx, chatManage.KnowledgeBaseID, searchParams)
logger.Infof(ctx, "Search by processed query: %s, results count: %d, error: %v",
searchParams.QueryText, len(searchResults), err,
)
if err != nil {
return ErrSearch.WithError(err)
logger.Infof(ctx, "Searching with processed query: %s", searchParams.QueryText)
var wg2 sync.WaitGroup
var mu2 sync.Mutex
var processedResults []*types.SearchResult
for _, kbID := range knowledgeBaseIDs {
wg2.Add(1)
go func(knowledgeBaseID string) {
defer wg2.Done()
results, err := p.knowledgeBaseService.HybridSearch(ctx, knowledgeBaseID, searchParams)
if err != nil {
logger.Errorf(ctx, "Failed to search KB %s with processed query: %v", knowledgeBaseID, err)
return
}
logger.Infof(ctx, "KB %s processed query results count: %d", knowledgeBaseID, len(results))
mu2.Lock()
processedResults = append(processedResults, results...)
mu2.Unlock()
}(kbID)
}
chatManage.SearchResult = append(chatManage.SearchResult, searchResults...)
wg2.Wait()
logger.Infof(ctx, "Total processed query results from all KBs: %d", len(processedResults))
chatManage.SearchResult = append(chatManage.SearchResult, processedResults...)
}
// remove duplicate results
// Remove duplicate results
chatManage.SearchResult = removeDuplicateResults(chatManage.SearchResult)
// Return if we have results
@@ -2,6 +2,7 @@ package chatpipline
import (
"context"
"sync"
"github.com/Tencent/WeKnora/internal/logger"
"github.com/Tencent/WeKnora/internal/types"
@@ -45,17 +46,56 @@ func (p *PluginSearchEntity) OnEvent(ctx context.Context,
return next()
}
graph, err := p.graphRepo.SearchNode(ctx, types.NameSpace{KnowledgeBase: chatManage.KnowledgeBaseID}, entity)
if err != nil {
logger.Errorf(ctx, "Failed to search node, session_id: %s, error: %v", chatManage.SessionID, err)
// Get knowledge base IDs list
knowledgeBaseIDs := chatManage.KnowledgeBaseIDs
if len(knowledgeBaseIDs) == 0 && chatManage.KnowledgeBaseID != "" {
knowledgeBaseIDs = []string{chatManage.KnowledgeBaseID}
logger.Infof(ctx, "No KnowledgeBaseIDs provided, falling back to single KB: %s", chatManage.KnowledgeBaseID)
}
if len(knowledgeBaseIDs) == 0 {
logger.Warnf(ctx, "No knowledge base IDs available for entity search")
return next()
}
chatManage.GraphResult = graph
logger.Infof(ctx, "search entity result count: %d", len(graph.Node))
// graphStr, _ := json.Marshal(graph)
// logger.Debugf(ctx, "search entity result: %s", string(graphStr))
chunkIDs := filterSeenChunk(ctx, graph, chatManage.SearchResult)
logger.Infof(ctx, "Searching entities across %d knowledge base(s): %v", len(knowledgeBaseIDs), knowledgeBaseIDs)
// Parallel search across multiple knowledge bases
var wg sync.WaitGroup
var mu sync.Mutex
var allNodes []*types.GraphNode
var allRelations []*types.GraphRelation
for _, kbID := range knowledgeBaseIDs {
wg.Add(1)
go func(knowledgeBaseID string) {
defer wg.Done()
graph, err := p.graphRepo.SearchNode(ctx, types.NameSpace{KnowledgeBase: knowledgeBaseID}, entity)
if err != nil {
logger.Errorf(ctx, "Failed to search entity in KB %s: %v", knowledgeBaseID, err)
return
}
logger.Infof(ctx, "KB %s entity search result count: %d nodes, %d relations", knowledgeBaseID, len(graph.Node), len(graph.Relation))
mu.Lock()
allNodes = append(allNodes, graph.Node...)
allRelations = append(allRelations, graph.Relation...)
mu.Unlock()
}(kbID)
}
wg.Wait()
// Merge graph data
chatManage.GraphResult = &types.GraphData{
Node: allNodes,
Relation: allRelations,
}
logger.Infof(ctx, "Total entity search result: %d nodes, %d relations", len(allNodes), len(allRelations))
chunkIDs := filterSeenChunk(ctx, chatManage.GraphResult, chatManage.SearchResult)
if len(chunkIDs) == 0 {
logger.Infof(ctx, "No new chunk found")
return next()
@@ -1,354 +0,0 @@
package service
import (
"context"
"fmt"
"strings"
"sync"
"github.com/Tencent/WeKnora/internal/logger"
"github.com/Tencent/WeKnora/internal/models/chat"
"github.com/Tencent/WeKnora/internal/types/interfaces"
)
// sessionContext holds the context for a single session
type sessionContext struct {
Messages []chat.Message
mu sync.RWMutex
}
// contextManager implements the ContextManager interface
type contextManager struct {
sessions map[string]*sessionContext
mu sync.RWMutex
compressionStrategy interfaces.CompressionStrategy
maxTokens int // Maximum tokens allowed in context
}
// NewContextManager creates a new context manager
func NewContextManager(compressionStrategy interfaces.CompressionStrategy, maxTokens int) interfaces.ContextManager {
return &contextManager{
sessions: make(map[string]*sessionContext),
compressionStrategy: compressionStrategy,
maxTokens: maxTokens,
}
}
// getOrCreateSession gets or creates a session context
func (cm *contextManager) getOrCreateSession(sessionID string) *sessionContext {
cm.mu.Lock()
defer cm.mu.Unlock()
if session, exists := cm.sessions[sessionID]; exists {
return session
}
session := &sessionContext{
Messages: make([]chat.Message, 0),
}
cm.sessions[sessionID] = session
return session
}
// AddMessage adds a message to the session context
func (cm *contextManager) AddMessage(ctx context.Context, sessionID string, message chat.Message) error {
logger.Infof(ctx, "[ContextManager][Session-%s] Adding message: role=%s, content_length=%d",
sessionID, message.Role, len(message.Content))
logger.Debugf(ctx, "Adding message to context, session: %s, role: %s", sessionID, message.Role)
// Log message content preview
contentPreview := message.Content
if len(contentPreview) > 200 {
contentPreview = contentPreview[:200] + "..."
}
logger.Debugf(ctx, "[ContextManager][Session-%s] Message content preview: %s", sessionID, contentPreview)
session := cm.getOrCreateSession(sessionID)
session.mu.Lock()
defer session.mu.Unlock()
beforeCount := len(session.Messages)
session.Messages = append(session.Messages, message)
logger.Debugf(ctx, "[ContextManager][Session-%s] Messages count: %d -> %d", sessionID, beforeCount, len(session.Messages))
// Check if compression is needed
tokenCount := cm.compressionStrategy.EstimateTokens(session.Messages)
logger.Debugf(ctx, "[ContextManager][Session-%s] Current token count: %d (max: %d)",
sessionID, tokenCount, cm.maxTokens)
if tokenCount > cm.maxTokens {
logger.Infof(ctx, "[ContextManager][Session-%s] Context exceeds max tokens (%d > %d), applying compression",
sessionID, tokenCount, cm.maxTokens)
beforeCompressionCount := len(session.Messages)
compressed, err := cm.compressionStrategy.Compress(ctx, session.Messages, cm.maxTokens)
if err != nil {
logger.Errorf(ctx, "[ContextManager][Session-%s] Failed to compress context: %v", sessionID, err)
return fmt.Errorf("failed to compress context: %w", err)
}
session.Messages = compressed
afterTokenCount := cm.compressionStrategy.EstimateTokens(session.Messages)
logger.Infof(ctx, "[ContextManager][Session-%s] Context compressed: %d -> %d messages, %d -> %d tokens",
sessionID, beforeCompressionCount, len(compressed), tokenCount, afterTokenCount)
}
return nil
}
// GetContext retrieves the current context for a session
func (cm *contextManager) GetContext(ctx context.Context, sessionID string) ([]chat.Message, error) {
logger.Infof(ctx, "[ContextManager][Session-%s] Getting context", sessionID)
logger.Debugf(ctx, "Getting context for session: %s", sessionID)
session := cm.getOrCreateSession(sessionID)
session.mu.RLock()
defer session.mu.RUnlock()
// Return a copy to avoid external modifications
messages := make([]chat.Message, len(session.Messages))
copy(messages, session.Messages)
// Calculate token estimate
tokenCount := cm.compressionStrategy.EstimateTokens(messages)
logger.Infof(ctx, "[ContextManager][Session-%s] Retrieved %d messages (~%d tokens)",
sessionID, len(messages), tokenCount)
logger.Debugf(ctx, "Retrieved %d messages from context", len(messages))
// Log message role distribution
roleCount := make(map[string]int)
for _, msg := range messages {
roleCount[msg.Role]++
}
logger.Debugf(ctx, "[ContextManager][Session-%s] Message distribution: %v", sessionID, roleCount)
return messages, nil
}
// ClearContext clears all context for a session
func (cm *contextManager) ClearContext(ctx context.Context, sessionID string) error {
logger.Infof(ctx, "[ContextManager][Session-%s] Clearing context", sessionID)
cm.mu.Lock()
defer cm.mu.Unlock()
// Check if session exists before deleting
if session, exists := cm.sessions[sessionID]; exists {
messageCount := len(session.Messages)
logger.Infof(ctx, "[ContextManager][Session-%s] Deleted %d messages", sessionID, messageCount)
} else {
logger.Debugf(ctx, "[ContextManager][Session-%s] Session not found (already cleared or never created)",
sessionID)
}
delete(cm.sessions, sessionID)
return nil
}
// GetContextStats returns statistics about the context
func (cm *contextManager) GetContextStats(ctx context.Context, sessionID string) (*interfaces.ContextStats, error) {
session := cm.getOrCreateSession(sessionID)
session.mu.RLock()
defer session.mu.RUnlock()
tokenCount := cm.compressionStrategy.EstimateTokens(session.Messages)
stats := &interfaces.ContextStats{
MessageCount: len(session.Messages),
TokenCount: tokenCount,
IsCompressed: false, // We'd need to track this explicitly for accurate reporting
OriginalMessageCount: len(session.Messages),
}
logger.Debugf(ctx, "Context stats for session %s: %d messages, ~%d tokens",
sessionID, stats.MessageCount, stats.TokenCount)
return stats, nil
}
// SlidingWindowStrategy implements a sliding window compression strategy
// It keeps the most recent N messages and discards older ones
type SlidingWindowStrategy struct {
windowSize int // Number of messages to keep
}
// NewSlidingWindowStrategy creates a new sliding window strategy
func NewSlidingWindowStrategy(windowSize int) interfaces.CompressionStrategy {
return &SlidingWindowStrategy{
windowSize: windowSize,
}
}
// Compress applies sliding window compression
func (s *SlidingWindowStrategy) Compress(ctx context.Context, messages []chat.Message, maxTokens int) ([]chat.Message, error) {
if len(messages) <= s.windowSize {
return messages, nil
}
// Keep system messages and recent messages
var compressed []chat.Message
var systemMessages []chat.Message
var otherMessages []chat.Message
// Separate system messages
for _, msg := range messages {
if msg.Role == "system" {
systemMessages = append(systemMessages, msg)
} else {
otherMessages = append(otherMessages, msg)
}
}
// Keep last N non-system messages
startIdx := 0
if len(otherMessages) > s.windowSize {
startIdx = len(otherMessages) - s.windowSize
}
// Combine system messages with recent messages
compressed = append(compressed, systemMessages...)
compressed = append(compressed, otherMessages[startIdx:]...)
logger.Infof(ctx, "Sliding window compression: %d -> %d messages (kept %d system, %d recent)",
len(messages), len(compressed), len(systemMessages), len(compressed)-len(systemMessages))
return compressed, nil
}
// EstimateTokens provides a rough token count estimation
// Uses approximately 4 characters per token as a heuristic
func (s *SlidingWindowStrategy) EstimateTokens(messages []chat.Message) int {
totalChars := 0
for _, msg := range messages {
totalChars += len(msg.Content)
}
// Rough estimate: ~4 characters per token
return totalChars / 4
}
// SmartCompressionStrategy implements intelligent context compression
// It uses summarization for older messages while keeping recent ones intact
type SmartCompressionStrategy struct {
recentMessageCount int // Number of recent messages to keep intact
chatModel chat.Chat // LLM for summarization
summarizeThreshold int // Minimum messages before summarization
}
// NewSmartCompressionStrategy creates a new smart compression strategy
func NewSmartCompressionStrategy(recentMessageCount int, chatModel chat.Chat, summarizeThreshold int) interfaces.CompressionStrategy {
return &SmartCompressionStrategy{
recentMessageCount: recentMessageCount,
chatModel: chatModel,
summarizeThreshold: summarizeThreshold,
}
}
// Compress applies smart compression with summarization
func (s *SmartCompressionStrategy) Compress(ctx context.Context, messages []chat.Message, maxTokens int) ([]chat.Message, error) {
logger.Infof(ctx, "[ContextManager][Compress] Smart compression started: %d messages, max_tokens=%d",
len(messages), maxTokens)
if len(messages) <= s.recentMessageCount {
logger.Infof(ctx, "[ContextManager][Compress] No compression needed (%d <= %d)",
len(messages), s.recentMessageCount)
return messages, nil
}
// Separate messages into system, old, and recent
var systemMessages []chat.Message
var oldMessages []chat.Message
var recentMessages []chat.Message
for i, msg := range messages {
if msg.Role == "system" {
systemMessages = append(systemMessages, msg)
} else if i < len(messages)-s.recentMessageCount {
oldMessages = append(oldMessages, msg)
} else {
recentMessages = append(recentMessages, msg)
}
}
logger.Infof(ctx, "[ContextManager][Compress] Message distribution: system=%d, old=%d, recent=%d",
len(systemMessages), len(oldMessages), len(recentMessages))
// If we have enough old messages, summarize them
var compressed []chat.Message
if len(oldMessages) >= s.summarizeThreshold && s.chatModel != nil {
logger.Infof(ctx, "[ContextManager][Compress] Summarizing %d old messages (threshold: %d)",
len(oldMessages), s.summarizeThreshold)
// Build conversation text for summarization
var conversationText strings.Builder
for _, msg := range oldMessages {
conversationText.WriteString(fmt.Sprintf("%s: %s\n", msg.Role, msg.Content))
}
// Create summarization prompt
summaryPrompt := []chat.Message{
{
Role: "system",
Content: "You are a helpful assistant that summarizes conversation history. " +
"Summarize the following conversation concisely while preserving key information and context. " +
"Keep the summary under 200 words.",
},
{
Role: "user",
Content: conversationText.String(),
},
}
logger.Debugf(ctx, "[ContextManager][Compress] Conversation text length: %d chars",
conversationText.Len())
// Get summary from LLM
logger.Infof(ctx, "[ContextManager][Compress] Calling LLM for summarization...")
response, err := s.chatModel.Chat(ctx, summaryPrompt, &chat.ChatOptions{
Temperature: 0.3,
})
if err != nil {
logger.Warnf(ctx, "[ContextManager][Compress] Failed to summarize old messages: %v, falling back to truncation",
err)
// Fallback to keeping only recent messages
} else {
// Add summary as a system message
summaryContent := fmt.Sprintf("[Previous conversation summary]: %s", response.Content)
compressed = append(compressed, chat.Message{
Role: "system",
Content: summaryContent,
})
logger.Infof(ctx, "[ContextManager][Compress] Successfully summarized %d messages into %d chars",
len(oldMessages), len(summaryContent))
logger.Debugf(ctx, "[ContextManager][Compress] Summary: %s", response.Content)
}
} else {
// Not enough messages to summarize, keep them as is
logger.Infof(ctx, "[ContextManager][Compress] Not enough messages to summarize (%d < %d), keeping as is",
len(oldMessages), s.summarizeThreshold)
compressed = append(compressed, oldMessages...)
}
// Add system messages, compressed history, and recent messages
result := make([]chat.Message, 0, len(systemMessages)+len(compressed)+len(recentMessages))
result = append(result, systemMessages...)
result = append(result, compressed...)
result = append(result, recentMessages...)
beforeTokens := s.EstimateTokens(messages)
afterTokens := s.EstimateTokens(result)
logger.Infof(ctx, "[ContextManager][Compress] Smart compression completed: %d -> %d messages (system: %d, compressed: %d, recent: %d), tokens: %d -> %d",
len(messages), len(result), len(systemMessages), len(compressed), len(recentMessages),
beforeTokens, afterTokens)
return result, nil
}
// EstimateTokens provides a rough token count estimation
func (s *SmartCompressionStrategy) EstimateTokens(messages []chat.Message) int {
totalChars := 0
for _, msg := range messages {
totalChars += len(msg.Content)
}
// Rough estimate: ~4 characters per token
return totalChars / 4
}
@@ -0,0 +1,224 @@
package llmcontext
import (
"context"
"fmt"
"strings"
"github.com/Tencent/WeKnora/internal/logger"
"github.com/Tencent/WeKnora/internal/models/chat"
"github.com/Tencent/WeKnora/internal/types/interfaces"
)
// slidingWindowStrategy implements CompressionStrategy using sliding window
type slidingWindowStrategy struct {
recentMessageCount int
}
// NewSlidingWindowStrategy creates a new sliding window compression strategy
func NewSlidingWindowStrategy(recentMessageCount int) interfaces.CompressionStrategy {
return &slidingWindowStrategy{
recentMessageCount: recentMessageCount,
}
}
// Compress implements the sliding window compression
// Keeps system messages and the most recent N messages
func (s *slidingWindowStrategy) Compress(ctx context.Context, messages []chat.Message, maxTokens int) ([]chat.Message, error) {
if len(messages) <= s.recentMessageCount {
return messages, nil
}
// Separate system messages from regular messages
var systemMessages []chat.Message
var regularMessages []chat.Message
for _, msg := range messages {
if msg.Role == "system" {
systemMessages = append(systemMessages, msg)
} else {
regularMessages = append(regularMessages, msg)
}
}
// Keep the most recent N regular messages
var keptMessages []chat.Message
if len(regularMessages) > s.recentMessageCount {
keptMessages = regularMessages[len(regularMessages)-s.recentMessageCount:]
} else {
keptMessages = regularMessages
}
// Combine: system messages first, then recent messages
result := make([]chat.Message, 0, len(systemMessages)+len(keptMessages))
result = append(result, systemMessages...)
result = append(result, keptMessages...)
logger.Infof(ctx, "[SlidingWindow] Compressed %d messages to %d messages (kept %d recent + %d system)",
len(messages), len(result), len(keptMessages), len(systemMessages))
return result, nil
}
// EstimateTokens estimates token count (rough approximation: 4 characters ≈ 1 token)
func (s *slidingWindowStrategy) EstimateTokens(messages []chat.Message) int {
totalChars := 0
for _, msg := range messages {
totalChars += len(msg.Role) + len(msg.Content)
// Account for tool calls if present
if len(msg.ToolCalls) > 0 {
for _, tc := range msg.ToolCalls {
totalChars += len(tc.Function.Name) + len(tc.Function.Arguments)
}
}
}
return totalChars / 4 // Rough approximation
}
// smartCompressionStrategy implements CompressionStrategy using LLM summarization
type smartCompressionStrategy struct {
recentMessageCount int
chatModel chat.Chat
summarizeThreshold int // Minimum messages before summarization
}
// NewSmartCompressionStrategy creates a new smart compression strategy
func NewSmartCompressionStrategy(recentMessageCount int, chatModel chat.Chat, summarizeThreshold int) interfaces.CompressionStrategy {
return &smartCompressionStrategy{
recentMessageCount: recentMessageCount,
chatModel: chatModel,
summarizeThreshold: summarizeThreshold,
}
}
// Compress implements smart compression with LLM summarization
// Summarizes old messages and keeps recent messages intact
func (s *smartCompressionStrategy) Compress(ctx context.Context, messages []chat.Message, maxTokens int) ([]chat.Message, error) {
if len(messages) <= s.recentMessageCount {
return messages, nil
}
// Separate system messages, old messages, and recent messages
var systemMessages []chat.Message
var oldMessages []chat.Message
var recentMessages []chat.Message
systemCount := 0
for _, msg := range messages {
if msg.Role == "system" {
systemMessages = append(systemMessages, msg)
systemCount++
}
}
// Get regular messages (non-system)
regularMessages := make([]chat.Message, 0, len(messages)-systemCount)
for _, msg := range messages {
if msg.Role != "system" {
regularMessages = append(regularMessages, msg)
}
}
// Split regular messages into old and recent
if len(regularMessages) > s.recentMessageCount {
splitPoint := len(regularMessages) - s.recentMessageCount
oldMessages = regularMessages[:splitPoint]
recentMessages = regularMessages[splitPoint:]
} else {
recentMessages = regularMessages
}
// If old messages are few, no need to summarize
if len(oldMessages) < s.summarizeThreshold {
result := make([]chat.Message, 0, len(systemMessages)+len(regularMessages))
result = append(result, systemMessages...)
result = append(result, regularMessages...)
return result, nil
}
// Summarize old messages using LLM
summary, err := s.summarizeMessages(ctx, oldMessages)
if err != nil {
logger.Warnf(ctx, "[SmartCompression] Failed to summarize messages: %v, falling back to old messages", err)
// Fallback: return all messages if summarization fails
result := make([]chat.Message, 0, len(systemMessages)+len(regularMessages))
result = append(result, systemMessages...)
result = append(result, regularMessages...)
return result, nil
}
// Construct final message list: system + summary + recent
result := make([]chat.Message, 0, len(systemMessages)+1+len(recentMessages))
result = append(result, systemMessages...)
result = append(result, chat.Message{
Role: "system",
Content: fmt.Sprintf("Previous conversation summary:\n%s", summary),
})
result = append(result, recentMessages...)
logger.Infof(ctx, "[SmartCompression] Compressed %d messages to %d messages (summarized %d old + kept %d recent + %d system)",
len(messages), len(result), len(oldMessages), len(recentMessages), len(systemMessages))
return result, nil
}
// summarizeMessages uses LLM to create a summary of old messages
func (s *smartCompressionStrategy) summarizeMessages(ctx context.Context, messages []chat.Message) (string, error) {
// Build conversation text
var sb strings.Builder
for i, msg := range messages {
sb.WriteString(fmt.Sprintf("[%s] %s\n", msg.Role, msg.Content))
if i < len(messages)-1 {
sb.WriteString("\n")
}
}
// Create summarization prompt
summaryPrompt := []chat.Message{
{
Role: "system",
Content: "You are a helpful assistant that summarizes conversations. " +
"Provide a concise summary that captures the key points, decisions, and context. " +
"Keep the summary brief but informative.",
},
{
Role: "user",
Content: fmt.Sprintf("Please summarize the following conversation:\n\n%s", sb.String()),
},
}
// Call LLM for summarization
response, err := s.chatModel.Chat(ctx, summaryPrompt, &chat.ChatOptions{
Temperature: 0.3, // Lower temperature for more consistent summaries
MaxTokens: 500, // Limit summary length
})
if err != nil {
return "", fmt.Errorf("failed to generate summary: %w", err)
}
if response == nil || response.Content == "" {
return "", fmt.Errorf("no summary generated")
}
summary := response.Content
logger.Debugf(ctx, "[SmartCompression] Generated summary (%d chars) from %d messages",
len(summary), len(messages))
return summary, nil
}
// EstimateTokens estimates token count (rough approximation: 4 characters ≈ 1 token)
func (s *smartCompressionStrategy) EstimateTokens(messages []chat.Message) int {
totalChars := 0
for _, msg := range messages {
totalChars += len(msg.Role) + len(msg.Content)
// Account for tool calls if present
if len(msg.ToolCalls) > 0 {
for _, tc := range msg.ToolCalls {
totalChars += len(tc.Function.Name) + len(tc.Function.Arguments)
}
}
}
return totalChars / 4 // Rough approximation
}
@@ -0,0 +1,156 @@
package llmcontext
import (
"context"
"fmt"
"github.com/Tencent/WeKnora/internal/logger"
"github.com/Tencent/WeKnora/internal/models/chat"
"github.com/Tencent/WeKnora/internal/types/interfaces"
)
// contextManager implements the ContextManager interface
// It handles business logic (compression, token management) and delegates storage to ContextStorage
type contextManager struct {
storage ContextStorage // Storage backend (Redis, Memory, etc.)
compressionStrategy interfaces.CompressionStrategy // Compression strategy
maxTokens int // Maximum tokens allowed in context
}
// NewContextManager creates a new context manager with the specified storage and compression strategy
func NewContextManager(storage ContextStorage, compressionStrategy interfaces.CompressionStrategy, maxTokens int) interfaces.ContextManager {
return &contextManager{
storage: storage,
compressionStrategy: compressionStrategy,
maxTokens: maxTokens,
}
}
// NewContextManagerWithMemory creates a context manager with in-memory storage (for backward compatibility)
func NewContextManagerWithMemory(compressionStrategy interfaces.CompressionStrategy, maxTokens int) interfaces.ContextManager {
return &contextManager{
storage: NewMemoryStorage(),
compressionStrategy: compressionStrategy,
maxTokens: maxTokens,
}
}
// AddMessage adds a message to the session context
// This method handles the business logic: loading, appending, compression, and saving
func (cm *contextManager) AddMessage(ctx context.Context, sessionID string, message chat.Message) error {
logger.Infof(ctx, "[ContextManager][Session-%s] Adding message: role=%s, content_length=%d",
sessionID, message.Role, len(message.Content))
// Log message content preview
contentPreview := message.Content
if len(contentPreview) > 200 {
contentPreview = contentPreview[:200] + "..."
}
logger.Debugf(ctx, "[ContextManager][Session-%s] Message content preview: %s", sessionID, contentPreview)
// Load existing messages from storage
messages, err := cm.storage.Load(ctx, sessionID)
if err != nil {
logger.Errorf(ctx, "[ContextManager][Session-%s] Failed to load context: %v", sessionID, err)
return fmt.Errorf("failed to load context: %w", err)
}
// Add new message
beforeCount := len(messages)
messages = append(messages, message)
logger.Debugf(ctx, "[ContextManager][Session-%s] Messages count: %d -> %d", sessionID, beforeCount, len(messages))
// Check if compression is needed
tokenCount := cm.compressionStrategy.EstimateTokens(messages)
logger.Debugf(ctx, "[ContextManager][Session-%s] Current token count: %d (max: %d)",
sessionID, tokenCount, cm.maxTokens)
if tokenCount > cm.maxTokens {
logger.Infof(ctx, "[ContextManager][Session-%s] Context exceeds max tokens (%d > %d), applying compression",
sessionID, tokenCount, cm.maxTokens)
beforeCompressionCount := len(messages)
compressed, err := cm.compressionStrategy.Compress(ctx, messages, cm.maxTokens)
if err != nil {
logger.Errorf(ctx, "[ContextManager][Session-%s] Failed to compress context: %v", sessionID, err)
return fmt.Errorf("failed to compress context: %w", err)
}
messages = compressed
afterTokenCount := cm.compressionStrategy.EstimateTokens(messages)
logger.Infof(ctx, "[ContextManager][Session-%s] Context compressed: %d -> %d messages, %d -> %d tokens",
sessionID, beforeCompressionCount, len(compressed), tokenCount, afterTokenCount)
}
// Save updated messages to storage
if err := cm.storage.Save(ctx, sessionID, messages); err != nil {
logger.Errorf(ctx, "[ContextManager][Session-%s] Failed to save context: %v", sessionID, err)
return fmt.Errorf("failed to save context: %w", err)
}
logger.Infof(ctx, "[ContextManager][Session-%s] Successfully added message (total: %d messages)", sessionID, len(messages))
return nil
}
// GetContext retrieves the current context for a session from storage
func (cm *contextManager) GetContext(ctx context.Context, sessionID string) ([]chat.Message, error) {
logger.Infof(ctx, "[ContextManager][Session-%s] Getting context", sessionID)
// Load messages from storage
messages, err := cm.storage.Load(ctx, sessionID)
if err != nil {
logger.Errorf(ctx, "[ContextManager][Session-%s] Failed to load context: %v", sessionID, err)
return nil, fmt.Errorf("failed to load context: %w", err)
}
// Calculate token estimate
tokenCount := cm.compressionStrategy.EstimateTokens(messages)
logger.Infof(ctx, "[ContextManager][Session-%s] Retrieved %d messages (~%d tokens)",
sessionID, len(messages), tokenCount)
// Log message role distribution
roleCount := make(map[string]int)
for _, msg := range messages {
roleCount[msg.Role]++
}
logger.Debugf(ctx, "[ContextManager][Session-%s] Message distribution: %v", sessionID, roleCount)
return messages, nil
}
// ClearContext clears all context for a session from storage
func (cm *contextManager) ClearContext(ctx context.Context, sessionID string) error {
logger.Infof(ctx, "[ContextManager][Session-%s] Clearing context", sessionID)
// Delete from storage
if err := cm.storage.Delete(ctx, sessionID); err != nil {
logger.Errorf(ctx, "[ContextManager][Session-%s] Failed to clear context: %v", sessionID, err)
return fmt.Errorf("failed to clear context: %w", err)
}
logger.Infof(ctx, "[ContextManager][Session-%s] Context cleared successfully", sessionID)
return nil
}
// GetContextStats returns statistics about the context
func (cm *contextManager) GetContextStats(ctx context.Context, sessionID string) (*interfaces.ContextStats, error) {
// Load messages from storage
messages, err := cm.storage.Load(ctx, sessionID)
if err != nil {
logger.Errorf(ctx, "[ContextManager][Session-%s] Failed to load context for stats: %v", sessionID, err)
return nil, fmt.Errorf("failed to load context: %w", err)
}
tokenCount := cm.compressionStrategy.EstimateTokens(messages)
stats := &interfaces.ContextStats{
MessageCount: len(messages),
TokenCount: tokenCount,
IsCompressed: false, // We'd need to track this explicitly for accurate reporting
OriginalMessageCount: len(messages),
}
logger.Debugf(ctx, "[ContextManager][Session-%s] Context stats: %d messages, ~%d tokens",
sessionID, stats.MessageCount, stats.TokenCount)
return stats, nil
}
@@ -0,0 +1,74 @@
package llmcontext
import (
"context"
"github.com/Tencent/WeKnora/internal/logger"
"github.com/Tencent/WeKnora/internal/models/chat"
"github.com/Tencent/WeKnora/internal/types"
"github.com/Tencent/WeKnora/internal/types/interfaces"
)
const (
// Context manager types
ContextManagerTypeMemory = "memory"
ContextManagerTypeRedis = "redis"
// Default values
DefaultMaxTokens = 128 * 1024 // 128K tokens
DefaultRecentMessageCount = 20
DefaultSummarizeThreshold = 5
DefaultCompressionStrategy = "sliding_window"
)
// NewContextManagerFromConfig creates a ContextManager based on configuration
func NewContextManagerFromConfig(contextCfg *types.ContextConfig, storage ContextStorage, chatModel chat.Chat) interfaces.ContextManager {
// Use default values if config is nil
if contextCfg == nil {
logger.Info(context.TODO(), "ContextManager config not found, using default memory-based context manager")
strategy := NewSlidingWindowStrategy(DefaultRecentMessageCount)
storage := NewMemoryStorage()
return NewContextManager(storage, strategy, DefaultMaxTokens)
}
// Set default values if not specified
maxTokens := contextCfg.MaxTokens
if maxTokens == 0 {
maxTokens = DefaultMaxTokens
}
recentMessageCount := contextCfg.RecentMessageCount
if recentMessageCount == 0 {
recentMessageCount = DefaultRecentMessageCount
}
summarizeThreshold := contextCfg.SummarizeThreshold
if summarizeThreshold == 0 {
summarizeThreshold = DefaultSummarizeThreshold
}
compressionStrategy := contextCfg.CompressionStrategy
if compressionStrategy == "" {
compressionStrategy = DefaultCompressionStrategy
}
// Create compression strategy
var strategy interfaces.CompressionStrategy
switch compressionStrategy {
case "sliding_window":
strategy = NewSlidingWindowStrategy(recentMessageCount)
case "smart":
if chatModel != nil {
strategy = NewSmartCompressionStrategy(recentMessageCount, chatModel, summarizeThreshold)
} else {
logger.Warn(context.TODO(), "Smart compression requested but no chat model provided, falling back to sliding window")
strategy = NewSlidingWindowStrategy(recentMessageCount)
}
default:
logger.Warnf(context.TODO(), "Unknown compression strategy '%s', using sliding window", compressionStrategy)
strategy = NewSlidingWindowStrategy(recentMessageCount)
}
// Create context manager with storage and strategy
return NewContextManager(storage, strategy, maxTokens)
}
@@ -0,0 +1,65 @@
package llmcontext
import (
"context"
"sync"
"github.com/Tencent/WeKnora/internal/logger"
"github.com/Tencent/WeKnora/internal/models/chat"
)
// memoryStorage implements ContextStorage using in-memory storage
type memoryStorage struct {
sessions map[string][]chat.Message
mu sync.RWMutex
}
// NewMemoryStorage creates a new memory-based storage
func NewMemoryStorage() ContextStorage {
return &memoryStorage{
sessions: make(map[string][]chat.Message),
}
}
// Save saves messages for a session to memory
func (ms *memoryStorage) Save(ctx context.Context, sessionID string, messages []chat.Message) error {
ms.mu.Lock()
defer ms.mu.Unlock()
// Make a copy to avoid external modifications
messageCopy := make([]chat.Message, len(messages))
copy(messageCopy, messages)
ms.sessions[sessionID] = messageCopy
logger.Debugf(ctx, "[MemoryStorage][Session-%s] Saved %d messages to memory", sessionID, len(messages))
return nil
}
// Load loads messages for a session from memory
func (ms *memoryStorage) Load(ctx context.Context, sessionID string) ([]chat.Message, error) {
ms.mu.RLock()
defer ms.mu.RUnlock()
messages, exists := ms.sessions[sessionID]
if !exists {
logger.Debugf(ctx, "[MemoryStorage][Session-%s] No context found in memory", sessionID)
return []chat.Message{}, nil
}
// Return a copy to avoid external modifications
messageCopy := make([]chat.Message, len(messages))
copy(messageCopy, messages)
logger.Debugf(ctx, "[MemoryStorage][Session-%s] Loaded %d messages from memory", sessionID, len(messages))
return messageCopy, nil
}
// Delete deletes all messages for a session from memory
func (ms *memoryStorage) Delete(ctx context.Context, sessionID string) error {
ms.mu.Lock()
defer ms.mu.Unlock()
delete(ms.sessions, sessionID)
logger.Debugf(ctx, "[MemoryStorage][Session-%s] Deleted context from memory", sessionID)
return nil
}
@@ -0,0 +1,112 @@
package llmcontext
import (
"context"
"encoding/json"
"fmt"
"time"
"github.com/Tencent/WeKnora/internal/logger"
"github.com/Tencent/WeKnora/internal/models/chat"
"github.com/redis/go-redis/v9"
)
// redisStorage implements ContextStorage using Redis
type redisStorage struct {
client *redis.Client
ttl time.Duration
prefix string
}
// NewRedisStorage creates a new Redis-based storage
func NewRedisStorage(client *redis.Client, ttl time.Duration, prefix string) (ContextStorage, error) {
// Validate connection
_, err := client.Ping(context.Background()).Result()
if err != nil {
return nil, fmt.Errorf("failed to connect to Redis: %w", err)
}
if ttl == 0 {
ttl = 24 * time.Hour // Default TTL 24 hours
}
if prefix == "" {
prefix = "context:" // Default prefix
}
return &redisStorage{
client: client,
ttl: ttl,
prefix: prefix,
}, nil
}
// buildKey builds the Redis key for a session
func (rs *redisStorage) buildKey(sessionID string) string {
return fmt.Sprintf("%s%s", rs.prefix, sessionID)
}
// Save saves messages for a session to Redis
func (rs *redisStorage) Save(ctx context.Context, sessionID string, messages []chat.Message) error {
key := rs.buildKey(sessionID)
// Marshal messages to JSON
data, err := json.Marshal(messages)
if err != nil {
logger.Errorf(ctx, "[RedisStorage][Session-%s] Failed to marshal messages: %v", sessionID, err)
return fmt.Errorf("failed to marshal messages: %w", err)
}
// Save to Redis with TTL
err = rs.client.Set(ctx, key, data, rs.ttl).Err()
if err != nil {
logger.Errorf(ctx, "[RedisStorage][Session-%s] Failed to save to Redis: %v", sessionID, err)
return fmt.Errorf("failed to save to Redis: %w", err)
}
logger.Debugf(ctx, "[RedisStorage][Session-%s] Saved %d messages to Redis (TTL: %s)",
sessionID, len(messages), rs.ttl)
return nil
}
// Load loads messages for a session from Redis
func (rs *redisStorage) Load(ctx context.Context, sessionID string) ([]chat.Message, error) {
key := rs.buildKey(sessionID)
// Get from Redis
data, err := rs.client.Get(ctx, key).Bytes()
if err != nil {
if err == redis.Nil {
// No context exists yet, return empty slice
logger.Debugf(ctx, "[RedisStorage][Session-%s] No context found in Redis", sessionID)
return []chat.Message{}, nil
}
logger.Errorf(ctx, "[RedisStorage][Session-%s] Failed to get from Redis: %v", sessionID, err)
return nil, fmt.Errorf("failed to get from Redis: %w", err)
}
// Unmarshal messages
var messages []chat.Message
err = json.Unmarshal(data, &messages)
if err != nil {
logger.Errorf(ctx, "[RedisStorage][Session-%s] Failed to unmarshal messages: %v", sessionID, err)
return nil, fmt.Errorf("failed to unmarshal messages: %w", err)
}
logger.Debugf(ctx, "[RedisStorage][Session-%s] Loaded %d messages from Redis", sessionID, len(messages))
return messages, nil
}
// Delete deletes all messages for a session from Redis
func (rs *redisStorage) Delete(ctx context.Context, sessionID string) error {
key := rs.buildKey(sessionID)
err := rs.client.Del(ctx, key).Err()
if err != nil {
logger.Errorf(ctx, "[RedisStorage][Session-%s] Failed to delete from Redis: %v", sessionID, err)
return fmt.Errorf("failed to delete from Redis: %w", err)
}
logger.Debugf(ctx, "[RedisStorage][Session-%s] Deleted context from Redis", sessionID)
return nil
}
@@ -0,0 +1,20 @@
package llmcontext
import (
"context"
"github.com/Tencent/WeKnora/internal/models/chat"
)
// ContextStorage defines the interface for storing and retrieving conversation context
// This separates storage implementation from business logic
type ContextStorage interface {
// Save saves messages for a session
Save(ctx context.Context, sessionID string, messages []chat.Message) error
// Load loads messages for a session
Load(ctx context.Context, sessionID string) ([]chat.Message, error)
// Delete deletes all messages for a session
Delete(ctx context.Context, sessionID string) error
}
+204 -142
View File
@@ -7,6 +7,7 @@ import (
"strings"
chatpipline "github.com/Tencent/WeKnora/internal/application/service/chat_pipline"
llmcontext "github.com/Tencent/WeKnora/internal/application/service/llmcontext"
"github.com/Tencent/WeKnora/internal/config"
"github.com/Tencent/WeKnora/internal/event"
"github.com/Tencent/WeKnora/internal/logger"
@@ -28,7 +29,7 @@ type sessionService struct {
tenantService interfaces.TenantService // Service for tenant operations
eventManager *chatpipline.EventManager // Event manager for chat pipeline
agentService interfaces.AgentService // Service for agent operations
contextManager interfaces.ContextManager // LLM context manager (separate from message storage)
sessionStorage llmcontext.ContextStorage // Session storage
}
// NewSessionService creates a new session service instance with all required dependencies
@@ -40,12 +41,8 @@ func NewSessionService(cfg *config.Config,
tenantService interfaces.TenantService,
eventManager *chatpipline.EventManager,
agentService interfaces.AgentService,
sessionStorage llmcontext.ContextStorage,
) interfaces.SessionService {
// Create default context manager with sliding window strategy
// Default: 8192 tokens max, keep last 20 messages
defaultStrategy := NewSlidingWindowStrategy(20)
contextManager := NewContextManager(defaultStrategy, 8192)
return &sessionService{
cfg: cfg,
sessionRepo: sessionRepo,
@@ -55,7 +52,7 @@ func NewSessionService(cfg *config.Config,
tenantService: tenantService,
eventManager: eventManager,
agentService: agentService,
contextManager: contextManager,
sessionStorage: sessionStorage,
}
}
@@ -215,44 +212,29 @@ func (s *sessionService) DeleteSession(ctx context.Context, id string) error {
// GenerateTitle generates a title for the current conversation content
func (s *sessionService) GenerateTitle(ctx context.Context,
sessionID string, messages []types.Message,
session *types.Session, messages []types.Message,
) (string, error) {
logger.Info(ctx, "Start generating session title")
// Validate session ID
if sessionID == "" {
logger.Error(ctx, "Failed to generate title: session ID cannot be empty")
return "", errors.New("session id is required")
}
// Get tenant ID from context
tenantID := ctx.Value(types.TenantIDContextKey).(uint)
logger.Infof(ctx, "Getting session info, session ID: %s, tenant ID: %d", sessionID, tenantID)
// Get session from repository
session, err := s.sessionRepo.Get(ctx, tenantID, sessionID)
if err != nil {
logger.ErrorWithFields(ctx, err, map[string]interface{}{
"session_id": sessionID,
"tenant_id": tenantID,
})
return "", err
if session == nil {
logger.Error(ctx, "Failed to generate title: session cannot be empty")
return "", errors.New("session cannot be empty")
}
// Skip if title already exists
if session.Title != "" {
logger.Infof(ctx, "Session already has a title, session ID: %s, title: %s", sessionID, session.Title)
logger.Infof(ctx, "Session already has a title, session ID: %s, title: %s", session.ID, session.Title)
return session.Title, nil
}
var err error
// Get the first user message, either from provided messages or repository
var message *types.Message
if len(messages) == 0 {
logger.Info(ctx, "Message list is empty, getting the first user message")
message, err = s.messageRepo.GetFirstMessageOfUser(ctx, sessionID)
message, err = s.messageRepo.GetFirstMessageOfUser(ctx, session.ID)
if err != nil {
logger.ErrorWithFields(ctx, err, map[string]interface{}{
"session_id": sessionID,
"session_id": session.ID,
})
return "", err
}
@@ -272,12 +254,40 @@ func (s *sessionService) GenerateTitle(ctx context.Context,
return "", errors.New("no user message found")
}
// Get chat model
logger.Infof(ctx, "Getting chat model, model ID: %s", session.SummaryModelID)
chatModel, err := s.modelService.GetChatModel(ctx, session.SummaryModelID)
// Get chat model, use default if SummaryModelID is empty
modelID := session.SummaryModelID
if modelID == "" {
logger.Info(ctx, "Session SummaryModelID is empty, trying to get default chat model")
// Try to get default KnowledgeQA model
models, err := s.modelService.ListModels(ctx)
if err != nil {
logger.ErrorWithFields(ctx, err, nil)
return "", fmt.Errorf("failed to list models: %w", err)
}
// Find default KnowledgeQA model or first KnowledgeQA model
for _, model := range models {
if model.Type == types.ModelTypeKnowledgeQA {
if model.IsDefault {
modelID = model.ID
logger.Infof(ctx, "Using default KnowledgeQA model: %s", modelID)
break
} else if modelID == "" {
modelID = model.ID
logger.Infof(ctx, "Using first available KnowledgeQA model: %s", modelID)
}
}
}
if modelID == "" {
logger.Error(ctx, "No KnowledgeQA model found")
return "", errors.New("no KnowledgeQA model available for title generation")
}
}
logger.Infof(ctx, "Getting chat model, model ID: %s", modelID)
chatModel, err := s.modelService.GetChatModel(ctx, modelID)
if err != nil {
logger.ErrorWithFields(ctx, err, map[string]interface{}{
"model_id": session.SummaryModelID,
"model_id": modelID,
})
return "", err
}
@@ -316,41 +326,99 @@ func (s *sessionService) GenerateTitle(ctx context.Context,
return "", err
}
logger.Infof(ctx, "Session title updated successfully, ID: %s, title: %s", sessionID, session.Title)
logger.Infof(ctx, "Session title updated successfully, ID: %s, title: %s", session.ID, session.Title)
return session.Title, nil
}
// GenerateTitleAsync generates a title for the session asynchronously
// This method clones the session and generates the title in a goroutine
// It emits an event when the title is generated
func (s *sessionService) GenerateTitleAsync(ctx context.Context, session *types.Session, userQuery string, eventBus *event.EventBus) {
// Extract values from context before cloning
tenantID := ctx.Value(types.TenantIDContextKey)
requestID := ctx.Value(types.RequestIDContextKey)
go func() {
// Create new background context and copy values
bgCtx := context.Background()
if tenantID != nil {
bgCtx = context.WithValue(bgCtx, types.TenantIDContextKey, tenantID)
}
if requestID != nil {
bgCtx = context.WithValue(bgCtx, types.RequestIDContextKey, requestID)
}
logger.Info(bgCtx, "Starting async title generation")
// Skip if title already exists
if session.Title != "" {
logger.Infof(bgCtx, "Session already has a title, skipping generation, session ID: %s", session.ID)
return
}
// Generate title using the first user message
messages := []types.Message{
{
Role: "user",
Content: userQuery,
},
}
title, err := s.GenerateTitle(bgCtx, session, messages)
if err != nil {
logger.ErrorWithFields(bgCtx, err, map[string]interface{}{
"session_id": session.ID,
})
return
}
// Emit title update event - BUG FIX: use bgCtx instead of ctx
// The original ctx is from the HTTP request and may be cancelled by the time we get here
if eventBus != nil {
if err := eventBus.Emit(bgCtx, event.Event{
Type: event.EventSessionTitle,
SessionID: session.ID,
Data: event.SessionTitleData{
SessionID: session.ID,
Title: title,
},
}); err != nil {
logger.ErrorWithFields(bgCtx, err, map[string]interface{}{
"session_id": session.ID,
})
} else {
logger.Infof(bgCtx, "Title update event emitted successfully, session ID: %s, title: %s", session.ID, title)
}
}
}()
}
// KnowledgeQA performs knowledge base question answering with LLM summarization
func (s *sessionService) KnowledgeQA(ctx context.Context, sessionID, query string) (
func (s *sessionService) KnowledgeQA(ctx context.Context, session *types.Session, query string, knowledgeBaseIDs []string) (
[]*types.SearchResult, <-chan types.StreamResponse, error,
) {
logger.Info(ctx, "Start knowledge base question answering")
logger.Infof(ctx, "Knowledge base question answering parameters, session ID: %s, query: %s", sessionID, query)
logger.Infof(ctx, "Knowledge base question answering parameters, session ID: %s, query: %s", session.ID, query)
// Get tenant ID from context
tenantID := ctx.Value(types.TenantIDContextKey).(uint)
logger.Infof(ctx, "Getting session info, session ID: %s, tenant ID: %d", sessionID, tenantID)
// Get session information
session, err := s.sessionRepo.Get(ctx, tenantID, sessionID)
if err != nil {
logger.Errorf(ctx, "Failed to get session, session ID: %s, error: %v", sessionID, err)
return nil, nil, err
// If no knowledge base IDs provided, fall back to session's default
if len(knowledgeBaseIDs) == 0 {
if session.KnowledgeBaseID != "" {
knowledgeBaseIDs = []string{session.KnowledgeBaseID}
logger.Infof(ctx, "No knowledge base IDs provided, using session default: %s", session.KnowledgeBaseID)
} else {
logger.Warnf(ctx, "Session has no associated knowledge base, session ID: %s", session.ID)
return nil, nil, errors.New("session has no knowledge base")
}
}
// Validate knowledge base association
if session.KnowledgeBaseID == "" {
logger.Warnf(ctx, "Session has no associated knowledge base, session ID: %s", sessionID)
return nil, nil, errors.New("session has no knowledge base")
}
logger.Infof(ctx, "Using knowledge bases: %v", knowledgeBaseIDs)
// Create chat management object with session settings
logger.Infof(ctx, "Creating chat manage object, knowledge base ID: %s", session.KnowledgeBaseID)
logger.Infof(ctx, "Creating chat manage object, knowledge base IDs: %v", knowledgeBaseIDs)
chatManage := &types.ChatManage{
Query: query,
RewriteQuery: query,
SessionID: sessionID,
KnowledgeBaseID: session.KnowledgeBaseID,
SessionID: session.ID,
KnowledgeBaseID: knowledgeBaseIDs[0], // For backward compatibility, use first KB ID
KnowledgeBaseIDs: knowledgeBaseIDs, // Multi-KB support
VectorThreshold: session.VectorThreshold,
KeywordThreshold: session.KeywordThreshold,
EmbeddingTopK: session.EmbeddingTopK,
@@ -377,10 +445,10 @@ func (s *sessionService) KnowledgeQA(ctx context.Context, sessionID, query strin
// Start knowledge QA event processing
logger.Info(ctx, "Triggering knowledge base question answering event")
err = s.KnowledgeQAByEvent(ctx, chatManage, types.Pipline["rag_stream"])
err := s.KnowledgeQAByEvent(ctx, chatManage, types.Pipline["rag_stream"])
if err != nil {
logger.ErrorWithFields(ctx, err, map[string]interface{}{
"session_id": sessionID,
"session_id": session.ID,
"knowledge_base_id": session.KnowledgeBaseID,
})
return nil, nil, err
@@ -539,24 +607,42 @@ func (s *sessionService) AgentQA(ctx context.Context, session *types.Session, qu
tenantID := ctx.Value(types.TenantIDContextKey).(uint)
logger.Infof(ctx, "Start agent-based question answering, session ID: %s, tenant ID: %d, query: %s", sessionID, tenantID, query)
// Get effective agent configuration (session > tenant)
var agentConfig *types.AgentConfig
// Build effective agent configuration by merging session and tenant configs
// Session-level config: Enabled, KnowledgeBases (stored in session.AgentConfig)
// Tenant-level config: MaxIterations, Temperature, Models, Tools, etc. (from tenant.AgentConfig)
// Fall back to tenant-level agent config (global default)
tenantInfo := ctx.Value(types.TenantInfoContextKey).(*types.Tenant)
if tenantInfo.AgentConfig != nil && tenantInfo.AgentConfig.Enabled {
logger.Infof(ctx, "Using tenant-level agent config for tenant: %d", tenantInfo.ID)
agentConfig = tenantInfo.AgentConfig
}
// Check if agent is enabled (either from session or tenant)
if agentConfig == nil || !agentConfig.Enabled {
logger.Warnf(ctx, "Agent not enabled for session: %s (neither session nor tenant has agent config)", sessionID)
// Check if agent is enabled at session level
if session.AgentConfig == nil || !session.AgentConfig.Enabled {
logger.Warnf(ctx, "Agent not enabled for session: %s", sessionID)
return nil, errors.New("agent not enabled for this session")
}
// Set knowledge bases for agent if not already configured
// Priority: AgentConfig.KnowledgeBases > Session.KnowledgeBaseID > All tenant knowledge bases
// Check if tenant has agent configuration
if tenantInfo.AgentConfig == nil {
logger.Warnf(ctx, "Tenant %d has no agent configuration", tenantInfo.ID)
return nil, errors.New("tenant has no agent configuration")
}
// Create runtime AgentConfig by merging session and tenant configs
// Tenant config provides the runtime parameters (MaxIterations, Temperature, Tools, Models)
// Session config provides Enabled and KnowledgeBases
agentConfig := &types.AgentConfig{
Enabled: session.AgentConfig.Enabled,
MaxIterations: tenantInfo.AgentConfig.MaxIterations,
ReflectionEnabled: tenantInfo.AgentConfig.ReflectionEnabled,
AllowedTools: tenantInfo.AgentConfig.AllowedTools,
Temperature: tenantInfo.AgentConfig.Temperature,
ThinkingModelID: tenantInfo.AgentConfig.ThinkingModelID,
RerankModelID: tenantInfo.AgentConfig.RerankModelID,
KnowledgeBases: session.AgentConfig.KnowledgeBases, // Use session's knowledge bases
}
logger.Infof(ctx, "Merged agent config from tenant %d and session %s", tenantInfo.ID, sessionID)
// Determine knowledge bases for agent
// Priority: Session.AgentConfig.KnowledgeBases > Session.KnowledgeBaseID > All tenant knowledge bases
if len(agentConfig.KnowledgeBases) == 0 {
if session.KnowledgeBaseID != "" {
// Use session's knowledge base as fallback
@@ -588,33 +674,42 @@ func (s *sessionService) AgentQA(ctx context.Context, session *types.Session, qu
logger.Infof(ctx, "Agent configured with %d knowledge base(s): %v", len(agentConfig.KnowledgeBases), agentConfig.KnowledgeBases)
}
// Set ThinkingModelID from session's SummaryModelID if not already set
// Set ThinkingModelID from session's SummaryModelID if not already set in tenant config
if agentConfig.ThinkingModelID == "" && session.SummaryModelID != "" {
agentConfig.ThinkingModelID = session.SummaryModelID
logger.Infof(ctx, "Using session's SummaryModelID as ThinkingModelID: %s", session.SummaryModelID)
}
// Create agent engine with EventBus
logger.Info(ctx, "Creating agent engine")
engine, err := s.agentService.CreateAgentEngine(ctx, agentConfig, eventBus)
// Get chat model
summaryModel, err := s.modelService.GetChatModel(ctx, agentConfig.ThinkingModelID)
if err != nil {
logger.Errorf(ctx, "Failed to create agent engine: %v", err)
return nil, err
logger.Warnf(ctx, "Failed to get chat model: %v", err)
return nil, fmt.Errorf("failed to get chat model: %w", err)
}
// Get or create contextManager for this session
contextManager := s.getContextManagerForSession(ctx, session, summaryModel)
// Get LLM context from context manager
llmContext, err := s.getContextForSession(ctx, session, sessionID)
llmContext, err := s.getContextForSession(ctx, contextManager, sessionID)
if err != nil {
logger.Warnf(ctx, "Failed to get LLM context: %v, continuing without history", err)
llmContext = []chat.Message{}
}
logger.Infof(ctx, "Loaded %d messages from LLM context manager", len(llmContext))
// Create agent engine with EventBus and ContextManager
logger.Info(ctx, "Creating agent engine")
engine, err := s.agentService.CreateAgentEngine(ctx, agentConfig, eventBus, contextManager, session.ID)
if err != nil {
logger.Errorf(ctx, "Failed to create agent engine: %v", err)
return nil, err
}
// Execute agent with streaming (asynchronously)
// Events will be emitted to EventBus and handled by the Handler layer
logger.Info(ctx, "Executing agent with streaming")
go func() {
if _, err := engine.Execute(ctx, sessionID, query, llmContext); err != nil {
if _, err := engine.Execute(ctx, sessionID, assistantMessageID, query, llmContext); err != nil {
logger.Errorf(ctx, "Agent execution failed: %v", err)
// Emit error event to the EventBus used by this agent
eventBus.Emit(ctx, event.Event{
@@ -633,44 +728,36 @@ func (s *sessionService) AgentQA(ctx context.Context, session *types.Session, qu
return nil, nil
}
// getContextForSession retrieves LLM context for a session
// Uses context manager which handles token limits and compression automatically
// This is separate from the message storage/conversation history
func (s *sessionService) getContextForSession(ctx context.Context, session *types.Session, sessionID string) ([]chat.Message, error) {
// Check if session has custom context configuration
var contextManager interfaces.ContextManager
if session.ContextConfig != nil && session.ContextConfig.Enabled {
// Create custom context manager based on session configuration
logger.Infof(ctx, "Using custom context config for session %s: strategy=%s, max_tokens=%d, recent_count=%d",
sessionID, session.ContextConfig.CompressionStrategy, session.ContextConfig.MaxTokens, session.ContextConfig.RecentMessageCount)
var strategy interfaces.CompressionStrategy
switch session.ContextConfig.CompressionStrategy {
case types.ContextCompressionSlidingWindow:
strategy = NewSlidingWindowStrategy(session.ContextConfig.RecentMessageCount)
case types.ContextCompressionSmart:
// For smart compression, we need the chat model
chatModel, err := s.modelService.GetChatModel(ctx, session.SummaryModelID)
if err != nil {
logger.Warnf(ctx, "Failed to get chat model for smart compression, falling back to sliding window: %v", err)
strategy = NewSlidingWindowStrategy(session.ContextConfig.RecentMessageCount)
} else {
strategy = NewSmartCompressionStrategy(session.ContextConfig.RecentMessageCount, chatModel, 5)
}
default:
logger.Warnf(ctx, "Unknown compression strategy %s, using sliding window", session.ContextConfig.CompressionStrategy)
strategy = NewSlidingWindowStrategy(session.ContextConfig.RecentMessageCount)
}
contextManager = NewContextManager(strategy, session.ContextConfig.MaxTokens)
// getContextManagerForSession creates a context manager for the session based on configuration
// Returns the configured context manager (tenant-level or session-level) or default
func (s *sessionService) getContextManagerForSession(ctx context.Context, session *types.Session, chatModel chat.Chat) interfaces.ContextManager {
// Get tenant to access global context configuration
tenant, _ := ctx.Value(types.TenantInfoContextKey).(*types.Tenant)
// Determine which context config to use: session-specific or tenant-level
var contextConfig *types.ContextConfig
if session.ContextConfig != nil {
// Use session-specific configuration
contextConfig = session.ContextConfig
logger.Infof(ctx, "Using session-specific context config for session %s", session.ID)
} else if tenant.ContextConfig != nil {
// Use tenant-level configuration
contextConfig = tenant.ContextConfig
logger.Infof(ctx, "Using tenant-level context config for session %s", session.ID)
} else {
// Use default context manager
logger.Debugf(ctx, "Using default context manager for session %s", sessionID)
contextManager = s.contextManager
// Use service's default context manager
logger.Debugf(ctx, "Using default context manager for session %s", session.ID)
contextConfig = &types.ContextConfig{
MaxTokens: llmcontext.DefaultMaxTokens,
CompressionStrategy: llmcontext.DefaultCompressionStrategy,
RecentMessageCount: llmcontext.DefaultRecentMessageCount,
SummarizeThreshold: llmcontext.DefaultSummarizeThreshold,
}
}
return llmcontext.NewContextManagerFromConfig(contextConfig, s.sessionStorage, chatModel)
}
// Get context from the context manager
// getContextForSession retrieves LLM context for a session
func (s *sessionService) getContextForSession(ctx context.Context, contextManager interfaces.ContextManager, sessionID string) ([]chat.Message, error) {
history, err := contextManager.GetContext(ctx, sessionID)
if err != nil {
return nil, fmt.Errorf("failed to get context: %w", err)
@@ -686,36 +773,11 @@ func (s *sessionService) getContextForSession(ctx context.Context, session *type
return history, nil
}
// AddMessageToContext adds a message to the LLM context
// This should be called after saving a message to the database
// The context manager handles token limits and compression automatically
func (s *sessionService) AddMessageToContext(ctx context.Context, session *types.Session, sessionID string, message chat.Message) error {
// Determine which context manager to use
var contextManager interfaces.ContextManager
if session.ContextConfig != nil && session.ContextConfig.Enabled {
// Create custom context manager based on session configuration
var strategy interfaces.CompressionStrategy
switch session.ContextConfig.CompressionStrategy {
case types.ContextCompressionSlidingWindow:
strategy = NewSlidingWindowStrategy(session.ContextConfig.RecentMessageCount)
case types.ContextCompressionSmart:
chatModel, err := s.modelService.GetChatModel(ctx, session.SummaryModelID)
if err != nil {
strategy = NewSlidingWindowStrategy(session.ContextConfig.RecentMessageCount)
} else {
strategy = NewSmartCompressionStrategy(session.ContextConfig.RecentMessageCount, chatModel, 5)
}
default:
strategy = NewSlidingWindowStrategy(session.ContextConfig.RecentMessageCount)
}
contextManager = NewContextManager(strategy, session.ContextConfig.MaxTokens)
} else {
contextManager = s.contextManager
}
// Add message to context
return contextManager.AddMessage(ctx, sessionID, message)
// ClearContext clears the LLM context for a session
// This is useful when switching knowledge bases or agent modes to prevent context contamination
func (s *sessionService) ClearContext(ctx context.Context, sessionID string) error {
logger.Infof(ctx, "Clearing context for session: %s", sessionID)
return s.sessionStorage.Delete(ctx, sessionID)
}
// processAgentEvents is no longer needed - events are handled directly by Handler layer via EventBus subscription
+33
View File
@@ -16,6 +16,7 @@ import (
"github.com/elastic/go-elasticsearch/v8"
"github.com/neo4j/neo4j-go-driver/v6/neo4j"
"github.com/panjf2000/ants/v2"
"github.com/redis/go-redis/v9"
"go.uber.org/dig"
"gorm.io/driver/postgres"
"gorm.io/gorm"
@@ -28,6 +29,7 @@ import (
"github.com/Tencent/WeKnora/internal/application/service"
chatpipline "github.com/Tencent/WeKnora/internal/application/service/chat_pipline"
"github.com/Tencent/WeKnora/internal/application/service/file"
"github.com/Tencent/WeKnora/internal/application/service/llmcontext"
"github.com/Tencent/WeKnora/internal/application/service/retriever"
"github.com/Tencent/WeKnora/internal/config"
"github.com/Tencent/WeKnora/internal/event"
@@ -60,7 +62,9 @@ func BuildContainer(container *dig.Container) *dig.Container {
must(container.Provide(initTracer))
must(container.Provide(initDatabase))
must(container.Provide(initFileService))
must(container.Provide(initRedisClient))
must(container.Provide(initAntsPool))
must(container.Provide(initContextStorage))
// Register goroutine pool cleanup handler
must(container.Invoke(registerPoolCleanup))
@@ -166,6 +170,35 @@ func initTracer() (*tracing.Tracer, error) {
return tracing.InitTracer()
}
func initRedisClient() (*redis.Client, error) {
db, err := strconv.Atoi(os.Getenv("REDIS_DB"))
if err != nil {
return nil, err
}
client := redis.NewClient(&redis.Options{
Addr: os.Getenv("REDIS_ADDR"),
Password: os.Getenv("REDIS_PASSWORD"),
DB: db,
})
// 验证连接
_, err = client.Ping(context.Background()).Result()
if err != nil {
return nil, fmt.Errorf("连接Redis失败: %w", err)
}
return client, nil
}
func initContextStorage(redisClient *redis.Client) (llmcontext.ContextStorage, error) {
storage, err := llmcontext.NewRedisStorage(redisClient, 24*time.Hour, "context:")
if err != nil {
return nil, err
}
return storage, nil
}
// initDatabase initializes database connection
// Creates and configures database connection based on environment configuration
// Supports multiple database backends (PostgreSQL)
+3
View File
@@ -56,6 +56,9 @@ const (
// Error events
EventError EventType = "error" // 错误事件
// Session events
EventSessionTitle EventType = "session_title" // 会话标题更新
)
// Event represents an event in the system
+6
View File
@@ -195,3 +195,9 @@ type AgentReflectionData struct {
Iteration int `json:"iteration"`
Done bool `json:"done"` // Whether streaming is complete
}
// SessionTitleData represents session title update data
type SessionTitleData struct {
SessionID string `json:"session_id"`
Title string `json:"title"`
}
+316 -31
View File
@@ -75,21 +75,20 @@ type SessionStrategy struct {
}
// CreateSessionRequest represents a request to create a new session
// Sessions are now knowledge-base-independent and serve as conversation containers.
// Knowledge bases can be specified dynamically in each query request (AgentQA/KnowledgeQA).
type CreateSessionRequest struct {
// ID of the associated knowledge base (optional in agent mode)
// ID of the associated knowledge base (optional, can be set/changed during queries)
KnowledgeBaseID string `json:"knowledge_base_id"`
// Session strategy configuration
SessionStrategy *SessionStrategy `json:"session_strategy"`
// Agent configuration (optional, for agent mode)
AgentConfig *types.AgentConfig `json:"agent_config"`
// Agent configuration (optional, session-level config only: enabled and knowledge_bases)
AgentConfig *types.SessionAgentConfig `json:"agent_config"`
}
// CreateSession handles the creation of a new conversation session
func (h *SessionHandler) CreateSession(c *gin.Context) {
ctx := c.Request.Context()
// logger.Infof(ctx, "Start creating session, config: %+v", h.config.Conversation)
// Parse and validate the request body
var request CreateSessionRequest
if err := c.ShouldBindJSON(&request); err != nil {
@@ -107,22 +106,14 @@ func (h *SessionHandler) CreateSession(c *gin.Context) {
}
// Validate session creation request
// - Traditional RAG mode: requires KnowledgeBaseID
// - Agent mode: KnowledgeBaseID is optional
// - If AgentConfig.KnowledgeBases is specified, use those
// - If session.KnowledgeBaseID is specified, use that
// - Otherwise, default to all knowledge bases under the tenant
// Sessions are now knowledge-base-independent:
// - KnowledgeBaseID is optional during session creation
// - Knowledge base can be specified in each query request (AgentQA/KnowledgeQA)
// - Agent mode can access multiple knowledge bases via AgentConfig.KnowledgeBases
// - Knowledge base can be switched during conversation
isAgentMode := request.AgentConfig != nil && request.AgentConfig.Enabled
hasKnowledgeBase := request.KnowledgeBaseID != ""
hasAgentKnowledgeBases := request.AgentConfig != nil && len(request.AgentConfig.KnowledgeBases) > 0
// In traditional RAG mode, knowledge base ID is required
if !isAgentMode && !hasKnowledgeBase {
logger.Error(ctx, "Traditional RAG mode requires knowledge_base_id")
c.Error(errors.NewBadRequestError("knowledge_base_id is required for non-agent sessions"))
return
}
logger.Infof(
ctx,
"Processing session creation request, tenant ID: %d, knowledge base ID: %s, agent mode: %v, agent KBs: %v",
@@ -208,8 +199,8 @@ func (h *SessionHandler) CreateSession(c *gin.Context) {
logger.Debug(ctx, "Using default session strategy")
}
// Only fetch knowledge base if KnowledgeBaseID is provided
// In agent mode, knowledge bases may be configured in AgentConfig instead
// Fetch knowledge base if KnowledgeBaseID is provided to inherit its model configurations
// If no KB is provided, models will be determined at query time or use tenant/system defaults
if request.KnowledgeBaseID != "" {
kb, err := h.knowledgebaseService.GetKnowledgeBaseByID(ctx, request.KnowledgeBaseID)
if err != nil {
@@ -229,7 +220,7 @@ func (h *SessionHandler) CreateSession(c *gin.Context) {
logger.Debugf(ctx, "Knowledge base fetched: %s, rerank model: %s, summary model: %s",
kb.ID, kb.RerankModelID, kb.SummaryModelID)
} else {
logger.Debug(ctx, "No knowledge base ID provided, skipping knowledge base fetch (agent mode)")
logger.Debug(ctx, "No knowledge base ID provided, models will use session strategy or be determined at query time")
}
// Call service to create session
@@ -437,9 +428,17 @@ func (h *SessionHandler) GenerateTitle(c *gin.Context) {
return
}
// Get session from database
session, err := h.sessionService.GetSession(ctx, sessionID)
if err != nil {
logger.ErrorWithFields(ctx, err, nil)
c.Error(errors.NewInternalServerError(err.Error()))
return
}
// Call service to generate title
logger.Infof(ctx, "Generating session title, session ID: %s, message count: %d", sessionID, len(request.Messages))
title, err := h.sessionService.GenerateTitle(ctx, sessionID, request.Messages)
title, err := h.sessionService.GenerateTitle(ctx, session, request.Messages)
if err != nil {
logger.ErrorWithFields(ctx, err, nil)
c.Error(errors.NewInternalServerError(err.Error()))
@@ -456,7 +455,9 @@ func (h *SessionHandler) GenerateTitle(c *gin.Context) {
// CreateKnowledgeQARequest defines the request structure for knowledge QA
type CreateKnowledgeQARequest struct {
Query string `json:"query" binding:"required"` // Query text for knowledge base search
Query string `json:"query" binding:"required"` // Query text for knowledge base search
KnowledgeBaseIDs []string `json:"knowledge_base_ids"` // Selected knowledge base ID for this request
AgentEnabled bool `json:"agent_enabled"` // Whether agent mode is enabled for this request
}
// SearchKnowledgeRequest defines the request structure for searching knowledge without LLM summarization
@@ -760,6 +761,44 @@ func (h *SessionHandler) KnowledgeQA(c *gin.Context) {
logger.Infof(ctx, "Knowledge QA request, session ID: %s, query: %s", sessionID, request.Query)
// Get request ID for title generation
requestID := c.GetString(types.RequestIDContextKey.String())
// Get session to check if title needs to be generated
session, err := h.sessionService.GetSession(ctx, sessionID)
if err != nil {
logger.Errorf(ctx, "Failed to get session, session ID: %s, error: %v", sessionID, err)
// Continue anyway - this is not critical
} else {
// Start async title generation if session has no title
if session.Title == "" {
logger.Infof(ctx, "Session has no title, starting async title generation, session ID: %s", sessionID)
// Use a simple event bus for title generation
titleEventBus := event.NewEventBus()
// Subscribe to title events and push to SSE
titleEventBus.On(event.EventSessionTitle, func(ctx context.Context, evt event.Event) error {
data, ok := evt.Data.(event.SessionTitleData)
if !ok {
return nil
}
// Send title update via SSE
c.SSEvent("message", &types.StreamResponse{
ID: requestID,
ResponseType: types.ResponseTypeSessionTitle,
Content: data.Title,
Done: true,
Data: map[string]interface{}{
"session_id": data.SessionID,
"title": data.Title,
},
})
c.Writer.Flush()
return nil
})
h.sessionService.GenerateTitleAsync(ctx, session, request.Query, titleEventBus)
}
}
// Create user message
if _, err := h.messageService.CreateMessage(ctx, &types.Message{
SessionID: sessionID,
@@ -781,8 +820,24 @@ func (h *SessionHandler) KnowledgeQA(c *gin.Context) {
}
logger.Infof(ctx, "Calling knowledge QA service, session ID: %s", sessionID)
// Prepare knowledge base IDs
knowledgeBaseIDs := request.KnowledgeBaseIDs
if len(knowledgeBaseIDs) == 0 && session.KnowledgeBaseID != "" {
knowledgeBaseIDs = []string{session.KnowledgeBaseID}
logger.Infof(ctx, "No knowledge base IDs in request, using session default: %s", session.KnowledgeBaseID)
}
// Validate knowledge bases
if len(knowledgeBaseIDs) == 0 {
logger.Error(ctx, "No knowledge base ID available")
c.Error(errors.NewBadRequestError("At least one knowledge base ID is required"))
return
}
logger.Infof(ctx, "Using knowledge bases: %v", knowledgeBaseIDs)
// Call service to perform knowledge QA
searchResults, respCh, err := h.sessionService.KnowledgeQA(ctx, sessionID, request.Query)
searchResults, respCh, err := h.sessionService.KnowledgeQA(ctx, session, request.Query, knowledgeBaseIDs)
if err != nil {
logger.ErrorWithFields(ctx, err, nil)
c.Error(errors.NewInternalServerError(err.Error()))
@@ -791,11 +846,16 @@ func (h *SessionHandler) KnowledgeQA(c *gin.Context) {
assistantMessage.KnowledgeReferences = searchResults
// Register new stream with stream manager
requestID := c.GetString(types.RequestIDContextKey.String())
if err := h.streamManager.RegisterStream(ctx, sessionID, assistantMessage.ID, request.Query); err != nil {
logger.GetLogger(ctx).Error("Register stream failed", "error", err)
}
// Set headers for SSE
c.Header("Content-Type", "text/event-stream")
c.Header("Cache-Control", "no-cache")
c.Header("Connection", "keep-alive")
c.Header("X-Accel-Buffering", "no")
// Send knowledge references if available
if len(searchResults) > 0 {
logger.Debugf(ctx, "Sending reference content, total %d", len(searchResults))
@@ -903,6 +963,177 @@ func (h *SessionHandler) AgentQA(c *gin.Context) {
return
}
// Initialize AgentConfig if it doesn't exist
if session.AgentConfig == nil {
session.AgentConfig = &types.SessionAgentConfig{}
}
// Detect if knowledge bases or agent mode has changed
knowledgeBasesChanged := false
configChanged := false
// Check if knowledge bases array has changed
if len(request.KnowledgeBaseIDs) > 0 {
// Compare arrays to detect changes
currentKBs := session.AgentConfig.KnowledgeBases
if len(currentKBs) != len(request.KnowledgeBaseIDs) {
knowledgeBasesChanged = true
configChanged = true
} else {
// Check if contents are different
kbMap := make(map[string]bool)
for _, kb := range currentKBs {
kbMap[kb] = true
}
for _, kb := range request.KnowledgeBaseIDs {
if !kbMap[kb] {
knowledgeBasesChanged = true
configChanged = true
break
}
}
}
if knowledgeBasesChanged {
logger.Infof(ctx, "Knowledge bases changed from %v to %v", session.AgentConfig.KnowledgeBases, request.KnowledgeBaseIDs)
}
}
// Check if agent mode has changed
currentAgentEnabled := session.AgentConfig.Enabled
if request.AgentEnabled != currentAgentEnabled {
logger.Infof(ctx, "Agent mode changed from %v to %v", currentAgentEnabled, request.AgentEnabled)
configChanged = true
}
// If configuration changed, clear context and update session
if configChanged {
logger.Infof(ctx, "Configuration changed, clearing context for session: %s", sessionID)
if knowledgeBasesChanged {
// Clear the LLM context to prevent contamination
if err := h.sessionService.ClearContext(ctx, sessionID); err != nil {
logger.Errorf(ctx, "Failed to clear context for session %s: %v", sessionID, err)
// Continue anyway - this is not a fatal error
}
}
session.AgentConfig.KnowledgeBases = request.KnowledgeBaseIDs
session.AgentConfig.Enabled = request.AgentEnabled
// Persist the session changes
if err := h.sessionService.UpdateSession(ctx, session); err != nil {
logger.Errorf(ctx, "Failed to update session %s: %v", sessionID, err)
c.Error(errors.NewInternalServerError("Failed to update session configuration"))
return
}
logger.Infof(ctx, "Session configuration updated successfully for session: %s", sessionID)
}
// If Agent mode is disabled, delegate to KnowledgeQA
if !request.AgentEnabled {
logger.Infof(ctx, "Agent mode disabled, delegating to KnowledgeQA for session: %s", sessionID)
// Use knowledge bases from request or session config
knowledgeBaseIDs := request.KnowledgeBaseIDs
if len(knowledgeBaseIDs) == 0 {
knowledgeBaseIDs = session.AgentConfig.KnowledgeBases
}
// If still empty, use session default knowledge base
if len(knowledgeBaseIDs) == 0 && session.KnowledgeBaseID != "" {
knowledgeBaseIDs = []string{session.KnowledgeBaseID}
logger.Infof(ctx, "Using session default knowledge base: %s", session.KnowledgeBaseID)
}
// Validate at least one knowledge base is available
if len(knowledgeBaseIDs) == 0 {
logger.Error(ctx, "No knowledge base available for delegation")
c.Error(errors.NewBadRequestError("No knowledge base available. Please configure at least one knowledge base."))
return
}
logger.Infof(ctx, "Delegating to KnowledgeQA with knowledge bases: %v", knowledgeBaseIDs)
// Create user message
requestID := c.GetString(types.RequestIDContextKey.String())
if _, err := h.messageService.CreateMessage(ctx, &types.Message{
SessionID: sessionID,
Role: "user",
Content: request.Query,
RequestID: requestID,
CreatedAt: time.Now(),
IsCompleted: true,
}); err != nil {
c.Error(errors.NewInternalServerError(err.Error()))
return
}
// Create assistant message (response)
assistantMessage.CreatedAt = time.Now()
if _, err := h.messageService.CreateMessage(ctx, assistantMessage); err != nil {
c.Error(errors.NewInternalServerError(err.Error()))
return
}
// Call KnowledgeQA service
searchResults, respCh, err := h.sessionService.KnowledgeQA(ctx, session, request.Query, knowledgeBaseIDs)
if err != nil {
logger.ErrorWithFields(ctx, err, nil)
c.Error(errors.NewInternalServerError(err.Error()))
return
}
assistantMessage.KnowledgeReferences = searchResults
// Register new stream with stream manager
if err := h.streamManager.RegisterStream(ctx, sessionID, assistantMessage.ID, request.Query); err != nil {
logger.GetLogger(ctx).Error("Register stream failed", "error", err)
}
// Set headers for SSE
c.Header("Content-Type", "text/event-stream")
c.Header("Cache-Control", "no-cache")
c.Header("Connection", "keep-alive")
c.Header("X-Accel-Buffering", "no")
// Send knowledge references if available
if len(searchResults) > 0 {
logger.Debugf(ctx, "Sending reference content, total %d", len(searchResults))
c.SSEvent("message", &types.StreamResponse{
ID: requestID,
ResponseType: types.ResponseTypeReferences,
KnowledgeReferences: searchResults,
})
c.Writer.Flush()
} else {
logger.Debug(ctx, "No reference content to send")
}
// Process streamed response
func() {
defer func() {
// Mark stream as completed when done
if err := h.streamManager.CompleteStream(ctx, sessionID, assistantMessage.ID); err != nil {
logger.GetLogger(ctx).Error("Complete stream failed", "error", err)
}
}()
for response := range respCh {
response.ID = requestID
c.SSEvent("message", response)
c.Writer.Flush()
if response.ResponseType == types.ResponseTypeAnswer {
assistantMessage.Content += response.Content
}
}
assistantMessage.IsCompleted = true
}()
// Update message with final content and references
assistantMessage.UpdatedAt = time.Now()
if err := h.messageService.UpdateMessage(ctx, assistantMessage); err != nil {
logger.Errorf(ctx, "Failed to update assistant message: %v", err)
}
logger.Infof(ctx, "KnowledgeQA delegation completed for session: %s", sessionID)
return
}
// Emit agent query event to create user message
requestID := c.GetString(types.RequestIDContextKey.String())
if err := event.Emit(ctx, event.Event{
@@ -916,6 +1147,38 @@ func (h *SessionHandler) AgentQA(c *gin.Context) {
},
}); err != nil {
logger.Errorf(ctx, "Failed to emit agent query event: %v", err)
return
}
// Set headers for SSE immediately
c.Header("Content-Type", "text/event-stream")
c.Header("Cache-Control", "no-cache")
c.Header("Connection", "keep-alive")
c.Header("X-Accel-Buffering", "no")
// Send agent query event to frontend via SSE to trigger loading state immediately
c.SSEvent("message", &types.StreamResponse{
ID: requestID,
ResponseType: types.ResponseTypeAgentQuery,
Content: "Agent query processing started",
Done: false,
Data: map[string]interface{}{
"session_id": sessionID,
"query": request.Query,
"request_id": requestID,
},
})
c.Writer.Flush()
// Create user message
if _, err := h.messageService.CreateMessage(ctx, &types.Message{
SessionID: sessionID,
Role: "user",
Content: request.Query,
RequestID: requestID,
CreatedAt: time.Now(),
IsCompleted: true,
}); err != nil {
c.Error(errors.NewInternalServerError(err.Error()))
return
}
@@ -936,11 +1199,7 @@ func (h *SessionHandler) AgentQA(c *gin.Context) {
logger.GetLogger(ctx).Error("Register stream failed", "error", err)
}
// Set headers for SSE
c.Header("Content-Type", "text/event-stream")
c.Header("Cache-Control", "no-cache")
c.Header("Connection", "keep-alive")
c.Header("X-Accel-Buffering", "no")
// SSE headers already set earlier when sending agent_query event
eventBus := event.NewEventBus()
// Create stream handler with dedicated EventBus BEFORE calling AgentQA
@@ -952,6 +1211,32 @@ func (h *SessionHandler) AgentQA(c *gin.Context) {
// Subscribe to events on the dedicated EventBus
streamHandler.Subscribe()
// Start async title generation if session has no title
if session.Title == "" {
logger.Infof(ctx, "Session has no title, starting async title generation, session ID: %s", sessionID)
// Subscribe to title events on the dedicated EventBus
eventBus.On(event.EventSessionTitle, func(ctx context.Context, evt event.Event) error {
data, ok := evt.Data.(event.SessionTitleData)
if !ok {
return nil
}
// Send title update via SSE
c.SSEvent("message", &types.StreamResponse{
ID: requestID,
ResponseType: types.ResponseTypeSessionTitle,
Content: data.Title,
Done: true,
Data: map[string]interface{}{
"session_id": data.SessionID,
"title": data.Title,
},
})
c.Writer.Flush()
return nil
})
h.sessionService.GenerateTitleAsync(ctx, session, request.Query, eventBus)
}
// Call service to perform agent QA
go func() {
searchResults, err := h.sessionService.AgentQA(ctx, session, request.Query, assistantMessage.ID, eventBus)
-1
View File
@@ -372,7 +372,6 @@ func (h *TenantHandler) UpdateTenantAgentConfig(c *gin.Context) {
Temperature: req.Temperature,
ThinkingModelID: req.ThinkingModelID,
RerankModelID: req.RerankModelID,
KnowledgeBases: []string{}, // Will be set per session
}
updatedTenant, err := h.service.UpdateTenant(ctx, tenant)
+26 -1
View File
@@ -7,7 +7,8 @@ import (
"time"
)
// AgentConfig represents the configuration for an agent session
// AgentConfig represents the full agent configuration (used at tenant level and runtime)
// This includes all configuration parameters for agent execution
type AgentConfig struct {
Enabled bool `json:"enabled"` // Whether agent mode is enabled
MaxIterations int `json:"max_iterations"` // Maximum number of ReAct iterations
@@ -19,6 +20,13 @@ type AgentConfig struct {
KnowledgeBases []string `json:"knowledge_bases"` // Accessible knowledge base IDs
}
// SessionAgentConfig represents session-level agent configuration
// Sessions only store Enabled and KnowledgeBases; other configs are read from Tenant at runtime
type SessionAgentConfig struct {
Enabled bool `json:"enabled"` // Whether agent mode is enabled for this session
KnowledgeBases []string `json:"knowledge_bases"` // Accessible knowledge base IDs for this session
}
// Value implements driver.Valuer interface for AgentConfig
func (c AgentConfig) Value() (driver.Value, error) {
return json.Marshal(c)
@@ -36,6 +44,23 @@ func (c *AgentConfig) Scan(value interface{}) error {
return json.Unmarshal(b, c)
}
// Value implements driver.Valuer interface for SessionAgentConfig
func (c SessionAgentConfig) Value() (driver.Value, error) {
return json.Marshal(c)
}
// Scan implements sql.Scanner interface for SessionAgentConfig
func (c *SessionAgentConfig) Scan(value interface{}) error {
if value == nil {
return nil
}
b, ok := value.([]byte)
if !ok {
return nil
}
return json.Unmarshal(b, c)
}
// Tool defines the interface that all agent tools must implement
type Tool interface {
// Name returns the unique identifier for this tool
+4
View File
@@ -54,6 +54,10 @@ const (
ResponseTypeError ResponseType = "error"
// Reflection response type (for agent reflection)
ResponseTypeReflection ResponseType = "reflection"
// Session title response type
ResponseTypeSessionTitle ResponseType = "session_title"
// Agent query response type (query received and processing started)
ResponseTypeAgentQuery ResponseType = "agent_query"
)
// StreamResponse stream response
+11 -5
View File
@@ -9,11 +9,12 @@ type ChatManage struct {
RewriteQuery string `json:"rewrite_query,omitempty"` // Query after rewriting for better retrieval
History []*History `json:"history,omitempty"` // Chat history for context
KnowledgeBaseID string `json:"knowledge_base_id"` // ID of the knowledge base to search against
VectorThreshold float64 `json:"vector_threshold"` // Minimum score threshold for vector search results
KeywordThreshold float64 `json:"keyword_threshold"` // Minimum score threshold for keyword search results
EmbeddingTopK int `json:"embedding_top_k"` // Number of top results to retrieve from embedding search
VectorDatabase string `json:"vector_database"` // Vector database type/name to use
KnowledgeBaseID string `json:"knowledge_base_id"` // ID of the knowledge base to search against (deprecated, use KnowledgeBaseIDs)
KnowledgeBaseIDs []string `json:"knowledge_base_ids"` // IDs of knowledge bases to search (multi-KB support)
VectorThreshold float64 `json:"vector_threshold"` // Minimum score threshold for vector search results
KeywordThreshold float64 `json:"keyword_threshold"` // Minimum score threshold for keyword search results
EmbeddingTopK int `json:"embedding_top_k"` // Number of top results to retrieve from embedding search
VectorDatabase string `json:"vector_database"` // Vector database type/name to use
RerankModelID string `json:"rerank_model_id"` // Model ID for reranking search results
RerankTopK int `json:"rerank_top_k"` // Number of top results after reranking
@@ -37,12 +38,17 @@ type ChatManage struct {
// Clone creates a deep copy of the ChatManage object
func (c *ChatManage) Clone() *ChatManage {
// Deep copy knowledge base IDs slice
knowledgeBaseIDs := make([]string, len(c.KnowledgeBaseIDs))
copy(knowledgeBaseIDs, c.KnowledgeBaseIDs)
return &ChatManage{
Query: c.Query,
ProcessedQuery: c.ProcessedQuery,
RewriteQuery: c.RewriteQuery,
SessionID: c.SessionID,
KnowledgeBaseID: c.KnowledgeBaseID,
KnowledgeBaseIDs: knowledgeBaseIDs,
VectorThreshold: c.VectorThreshold,
KeywordThreshold: c.KeywordThreshold,
EmbeddingTopK: c.EmbeddingTopK,
+3 -3
View File
@@ -20,13 +20,13 @@ type AgentStreamEvent struct {
// AgentEngine defines the interface for agent execution engine
type AgentEngine interface {
// ExecuteStreamWithHistory executes the agent with conversation history and returns a stream of events
Execute(ctx context.Context, sessionID, query string, llmContext []chat.Message) (*types.AgentState, error)
Execute(ctx context.Context, sessionID, messageID, query string, llmContext []chat.Message) (*types.AgentState, error)
}
// AgentService defines the interface for agent-related operations
type AgentService interface {
// CreateAgentEngine creates an agent engine with the given configuration and EventBus
CreateAgentEngine(ctx context.Context, config *types.AgentConfig, eventBus *event.EventBus) (AgentEngine, error)
// CreateAgentEngine creates an agent engine with the given configuration, EventBus, and ContextManager
CreateAgentEngine(ctx context.Context, config *types.AgentConfig, eventBus *event.EventBus, contextManager ContextManager, sessionID string) (AgentEngine, error)
// ValidateConfig validates an agent configuration
ValidateConfig(config *types.AgentConfig) error
+8 -2
View File
@@ -22,10 +22,14 @@ type SessionService interface {
// DeleteSession deletes a session
DeleteSession(ctx context.Context, id string) error
// GenerateTitle generates a title for the current conversation
GenerateTitle(ctx context.Context, sessionID string, messages []types.Message) (string, error)
GenerateTitle(ctx context.Context, session *types.Session, messages []types.Message) (string, error)
// GenerateTitleAsync generates a title for the session asynchronously
// It emits an event when the title is generated
GenerateTitleAsync(ctx context.Context, session *types.Session, userQuery string, eventBus *event.EventBus)
// KnowledgeQA performs knowledge-based question answering
// knowledgeBaseIDs: list of knowledge base IDs to search (supports multi-KB)
KnowledgeQA(ctx context.Context,
sessionID, query string,
session *types.Session, query string, knowledgeBaseIDs []string,
) ([]*types.SearchResult, <-chan types.StreamResponse, error)
// KnowledgeQAByEvent performs knowledge-based question answering by event
KnowledgeQAByEvent(ctx context.Context, chatManage *types.ChatManage, eventList []types.EventType) error
@@ -34,6 +38,8 @@ type SessionService interface {
// AgentQA performs agent-based question answering with conversation history and streaming support
// eventBus is optional - if nil, uses service's default EventBus
AgentQA(ctx context.Context, session *types.Session, query string, assistantMessageID string, eventBus *event.EventBus) ([]*types.SearchResult, error)
// ClearContext clears the LLM context for a session
ClearContext(ctx context.Context, sessionID string) error
}
// SessionRepository defines the session repository interface
+17 -17
View File
@@ -64,8 +64,8 @@ type ContextConfig struct {
// For sliding_window: number of messages to keep
// For smart: number of recent messages to keep uncompressed
RecentMessageCount int `json:"recent_message_count"`
// Enable context management (default: true)
Enabled bool `json:"enabled"`
// Summarize threshold: number of messages before summarization
SummarizeThreshold int `json:"summarize_threshold"`
}
// Session represents the session
@@ -80,21 +80,21 @@ type Session struct {
TenantID uint `json:"tenant_id" gorm:"index"`
// Strategy configuration
KnowledgeBaseID string `json:"knowledge_base_id"` // 关联的知识库ID
MaxRounds int `json:"max_rounds"` // 多轮保持轮数
EnableRewrite bool `json:"enable_rewrite"` // 多轮改写开关
FallbackStrategy FallbackStrategy `json:"fallback_strategy"` // 兜底策略
FallbackResponse string `json:"fallback_response"` // 固定回复内容
EmbeddingTopK int `json:"embedding_top_k"` // 向量召回TopK
KeywordThreshold float64 `json:"keyword_threshold"` // 关键词召回阈值
VectorThreshold float64 `json:"vector_threshold"` // 向量召回阈值
RerankModelID string `json:"rerank_model_id"` // 排序模型ID
RerankTopK int `json:"rerank_top_k"` // 排序TopK
RerankThreshold float64 `json:"rerank_threshold"` // 排序阈值
SummaryModelID string `json:"summary_model_id"` // 总结模型ID
SummaryParameters *SummaryConfig `json:"summary_parameters" gorm:"type:json"` // 总结模型参数
AgentConfig *AgentConfig `json:"agent_config" gorm:"type:json"` // Agent 配置(可选)
ContextConfig *ContextConfig `json:"context_config" gorm:"type:json"` // 上下文管理配置(可选)
KnowledgeBaseID string `json:"knowledge_base_id"` // 关联的知识库ID
MaxRounds int `json:"max_rounds"` // 多轮保持轮数
EnableRewrite bool `json:"enable_rewrite"` // 多轮改写开关
FallbackStrategy FallbackStrategy `json:"fallback_strategy"` // 兜底策略
FallbackResponse string `json:"fallback_response"` // 固定回复内容
EmbeddingTopK int `json:"embedding_top_k"` // 向量召回TopK
KeywordThreshold float64 `json:"keyword_threshold"` // 关键词召回阈值
VectorThreshold float64 `json:"vector_threshold"` // 向量召回阈值
RerankModelID string `json:"rerank_model_id"` // 排序模型ID
RerankTopK int `json:"rerank_top_k"` // 排序TopK
RerankThreshold float64 `json:"rerank_threshold"` // 排序阈值
SummaryModelID string `json:"summary_model_id"` // 总结模型ID
SummaryParameters *SummaryConfig `json:"summary_parameters" gorm:"type:json"` // 总结模型参数
AgentConfig *SessionAgentConfig `json:"agent_config" gorm:"type:json"` // Agent 配置(会话级别,仅存储enabled和knowledge_bases)
ContextConfig *ContextConfig `json:"context_config" gorm:"type:json"` // 上下文管理配置(可选)
CreatedAt time.Time `json:"created_at"`
UpdatedAt time.Time `json:"updated_at"`
+2
View File
@@ -30,6 +30,8 @@ type Tenant struct {
StorageUsed int64 `yaml:"storage_used" json:"storage_used" gorm:"default:0"`
// Global Agent configuration for this tenant (default for all sessions)
AgentConfig *AgentConfig `yaml:"agent_config" json:"agent_config" gorm:"type:json"`
// Global Context configuration for this tenant (default for all sessions)
ContextConfig *ContextConfig `yaml:"context_config" json:"context_config" gorm:"type:json"`
// Creation time
CreatedAt time.Time `yaml:"created_at" json:"created_at"`
// Last updated time