Complete phase two benchmarks, plot previews and static export workflow

2026-09-07 02:54:52 +08:00
parent 95095197df
commit 89df10bc4e
59 changed files with 2535 additions and 83 deletions
@@ -220,3 +220,8 @@ function close() { disarm(); viewer.value?.close(); svgHtml.value = ''; opener?.
 :is(.editor-mermaid-preview, .markdown-mermaid):is(:hover, :focus-within) > .diagram-controls { opacity: 1; pointer-events: auto; }
@media (hover: none) { :is(.editor-mermaid-preview, .markdown-mermaid) > .diagram-controls { opacity: 1; pointer-events: auto; } }
 </style>
+
+<style>
+/* Keep 10px axis labels readable on narrow screens; the existing container scrolls. */
+.function-plot-preview > svg, .markdown-function-plot > svg { min-width: 640px; }
+</style>
@@ -24,7 +24,7 @@ const diagramTheme = computed<'light' | 'dark'>(() => (themeStore.isDark ? 'dark
 // 主题切换需要重渲染：Mermaid SVG 的配色在渲染时烘焙，无法靠 CSS 变量事后调整。
 watch([() => props.source, diagramTheme, () => themeStore.currentThemeId, () => JSON.stringify(markdownPreferences.normalized), () => JSON.stringify([props.citationNumbers, props.citationAliases])], async ([source, theme]) => {
  const version = ++renderVersion
-  const result = await renderMarkdown(source, { theme, preferences: markdownPreferences.normalized, citationNumbers: props.citationNumbers, citationAliases: props.citationAliases })
+  const result = await renderMarkdown(source, { theme, themeId: themeStore.currentThemeId, preferences: markdownPreferences.normalized, citationNumbers: props.citationNumbers, citationAliases: props.citationAliases })
  if (version === renderVersion) html.value = result
 }, { immediate: true, flush: 'post' })
 </script>
@@ -20,6 +20,7 @@ const navItems = computed(() => [
  { name: 'plugins', icon: Connection, label: 'Plugin' },
  { name: 'mcp-servers', icon: Monitor, label: 'MCP' },
  { name: 'themes', icon: Brush, label: t('主题', 'Themes') },
+  { name: 'benchmarks', icon: Monitor, label: 'Benchmark' },
  { name: 'logs', icon: Document, label: t('日志', 'Logs') },
  { name: 'settings', icon: Setting, label: t('设置', 'Settings') },
 ])
@@ -0,0 +1,65 @@
+<script setup lang="ts">
+import { computed, ref, watch, onMounted, onBeforeUnmount } from 'vue'
+import { benchmarkService as service, type BenchmarkRun } from '@/services/benchmarkService'
+import { listProviders } from '@/services/providerService'
+const kind = ref<'rag' | 'agent'>('rag'), dataset = ref(''), error = ref(''), busy = ref(false)
+const datasets = ref<Awaited<ReturnType<typeof service.datasets>>>([]), runs = ref<BenchmarkRun[]>([])
+const providers = ref<Awaited<ReturnType<typeof listProviders>>>([]), provider = ref(''), model = ref('')
+const report = ref<Awaited<ReturnType<typeof service.report>> | null>(null)
+const topK = ref(5), rrfK = ref(60), rerank = ref(false)
+const fusion = ref<'rrf' | 'weighted'>('rrf')
+const metricLabels: Record<string, string> = {
+  total_cases: '计划样本', evaluated_cases: '已评样本', successful_cases: '运行成功', failed_cases: '运行失败',
+  task_success_rate: '任务成功率', tool_selection_accuracy: '工具选择准确率', tool_argument_accuracy: '参数准确率',
+  invalid_tool_call_rate: '无效调用率', average_steps: '平均步骤', average_latency_ms: '平均耗时 (ms)',
+  token_usage: 'Token 用量', tool_calls: '实际工具调用', expected_calls: '预期工具调用',
+  hit_at_1: 'Hit@1', hit_at_5: 'Hit@5', recall_at_k: 'Recall@K', mrr: 'MRR', citation_hit_rate: '引用命中率',
+  p50_latency_ms: 'P50 (ms)', p95_latency_ms: 'P95 (ms)', failure_rate: '运行失败率',
+}
+const metricGroups = computed(() => {
+  const metrics = report.value?.metrics ?? {}
+  const groups = 'task_success_rate' in metrics ? { Agent: metrics } : metrics
+  return Object.entries(groups).filter(([, value]) => value && typeof value === 'object').map(([name, value]) => ({
+    name, rows: Object.entries(value as Record<string, unknown>).map(([key, number]) => ({
+      label: metricLabels[key] ?? key,
+      value: number === null ? '不适用' : typeof number === 'number' ? Number(number.toFixed(4)).toLocaleString() : String(number),
+    })),
+  }))
+})
+let timer: ReturnType<typeof setTimeout> | undefined, disposed = false
+async function loadDatasets() { try { datasets.value = await service.datasets(kind.value); dataset.value = datasets.value[0]?.id ?? '' } catch(e) { error.value = String(e) } }
+async function refresh() { try { runs.value = await service.list() } catch(e) { error.value = String(e) } if (!disposed) timer = setTimeout(refresh, 1500) }
+watch(kind, loadDatasets)
+watch(provider, id => { model.value = providers.value.find(p => p.provider_id === id)?.default_model ?? '' })
+async function start() {
+  error.value = ''; busy.value = true
+  try { await service.start(kind.value, kind.value === 'agent' ? { dataset_id: dataset.value, provider_id: provider.value, model: model.value, max_steps: 6, timeout_seconds: 90, token_budget: 6000 } : { dataset_id: dataset.value, modes: ['fts','vector','hybrid'], retrieval: { top_k: topK.value, fusion: fusion.value, rrf_k: rrfK.value, rerank: rerank.value } }) }
+  catch(e) { error.value = String(e) } finally { busy.value = false }
+}
+async function action(run: BenchmarkRun, cancel = false) { try { if (cancel) await service.cancel(run.id); else report.value = await service.report(run.id) } catch(e) { error.value = String(e) } }
+function download() { const url = URL.createObjectURL(new Blob([JSON.stringify(report.value,null,2)], { type:'application/json' })); const a=document.createElement('a'); a.href=url; a.download='benchmark-report.json'; a.click(); setTimeout(()=>URL.revokeObjectURL(url),1000) }
+onMounted(async () => { void refresh(); void loadDatasets(); try { providers.value=(await listProviders()).filter(p=>p.enabled); provider.value=providers.value[0]?.provider_id ?? '' } catch(e) { error.value=String(e) } })
+onBeforeUnmount(() => { disposed=true; clearTimeout(timer) })
+</script>
+<template>
+  <main class="benchmark-page"><h1>Benchmark 评测</h1><p>标准数据集通过真实检索引擎或 Agent Runtime 执行。Agent 会使用所选提供商额度；需要权限时请打开 Trace 处理。</p>
+    <div class="controls"><label>类型 <select v-model="kind"><option value="rag">RAG</option><option value="agent">Agent</option></select></label>
+      <label>数据集 <select v-model="dataset"><option v-for="d in datasets" :key="d.id" :value="d.id">{{ d.id }} · {{ d.cases }} 案例</option></select></label>
+      <template v-if="kind === 'agent'"><label>提供商 <select v-model="provider"><option v-for="p in providers" :key="p.provider_id" :value="p.provider_id">{{ p.name }}</option></select></label><label>模型 <input v-model="model"></label></template>
+      <template v-else><label>融合 <select v-model="fusion"><option value="rrf">RRF</option><option value="weighted">加权 50/50</option></select></label><label>Top K <input v-model.number="topK" type="number" min="1" max="100"></label><label>RRF K <input v-model.number="rrfK" type="number" min="1"></label><label><input v-model="rerank" type="checkbox">Lexical Reranker</label></template>
+      <button :disabled="busy || !dataset || (kind === 'agent' && (!provider || !model))" @click="start">运行评测</button></div>
+    <p v-if="error" role="alert">{{ error }}</p>
+    <table><thead><tr><th>数据集</th><th>状态</th><th>操作</th></tr></thead><tbody><tr v-for="run in runs" :key="run.id"><td>{{ run.datasetId }}<small>{{ run.id }}</small></td><td>{{ run.status }} {{ run.progress === null ? '' : `${Math.round(run.progress*100)}%` }} {{ run.errorCode }}</td><td><button v-if="['queued','running'].includes(run.status)" @click="action(run,true)">取消</button><button v-else @click="action(run)">查看报告</button><RouterLink v-if="run.agentId" :to="`/agent/runs/${run.agentId}`">Agent Trace</RouterLink></td></tr></tbody></table>
+    <section v-if="report"><h2>评测报告</h2><button class="button-secondary" @click="download">下载完整 JSON</button>
+      <div v-for="group in metricGroups" :key="group.name"><h3>{{ group.name }}</h3><dl class="metric-grid"><div v-for="row in group.rows" :key="row.label"><dt>{{ row.label }}</dt><dd>{{ row.value }}</dd></div></dl></div>
+      <details><summary>冻结配置与逐例证据</summary><pre>{{ JSON.stringify(report,null,2) }}</pre></details></section>
+  </main>
+</template>
+<style scoped>
+.benchmark-page { padding:24px; overflow:auto; width:100%; } .controls { display:flex; gap:12px; flex-wrap:wrap; } label { display:flex; align-items:center; gap:6px; } input[type=number] { width:80px; } table { width:100%; margin-block:20px; border-collapse:collapse; } td,th { text-align:left; padding:12px; border-bottom:1px solid var(--color-border-default); } small { display:block; } pre { white-space:pre-wrap; overflow-wrap:anywhere; } [role=alert] { color:var(--color-error); }
+.metric-grid { display:grid; grid-template-columns:repeat(auto-fit,minmax(150px,1fr)); gap:12px; margin:16px 0; }
+.metric-grid > div { padding:16px; border:1px solid var(--color-border-default); border-radius:8px; background:var(--color-surface-primary); }
+dt { font-size:13px; color:var(--color-text-secondary); } dd { margin:8px 0 0; font-size:22px; font-weight:600; }
+button { padding:6px 12px; border:1px solid var(--color-border-default); border-radius:6px; background:var(--color-surface-primary); cursor:pointer; }
+button:disabled { opacity:.5; cursor:default; } td a { margin-left:12px; } input:not([type=checkbox]) { border:1px solid var(--color-border-default); border-radius:6px; padding:6px; }
+</style>
@@ -1,6 +1,7 @@
 <script setup lang="ts">
 import { useEditorStore } from '@/stores/editor'
 import { useWorkspaceStore } from '@/stores/workspace'
+import ExportDialog from './ExportDialog.vue'
 import { computed, ref } from 'vue'
 import ActionDialog from '@/components/common/ActionDialog.vue'
 import { useActionDialog } from '@/composables/useActionDialog'
@@ -10,6 +11,7 @@ const editorStore = useEditorStore()
 const workspaceStore = useWorkspaceStore()
 const { actionDialog, resolveAction, askConfirm } = useActionDialog()
 const reloadError = ref('')
+const exportOpen = ref(false)
 const needsRecovery = computed(() => ['conflict', 'external_changed'].includes(editorStore.saveStatus))
 const missingFile = computed(() => needsRecovery.value && editorStore.currentFilePath === workspaceStore.activeFilePath && !workspaceStore.activeFile && !workspaceStore.treeRefreshError)
 function downloadCopy() {
@@ -43,9 +45,11 @@ const statusText = computed<Record<string, string>>(() => ({

 <template>
  <header class="editor-header">
+    <ExportDialog v-if="exportOpen" @close="exportOpen = false" />
    <ActionDialog v-if="actionDialog" v-bind="actionDialog" @resolve="resolveAction" />
    <div class="file-identity"><strong>{{ workspaceStore.activeFile?.name ?? t('未命名笔记', 'Untitled note') }}</strong><small>{{ workspaceStore.activeFilePath }}</small></div>
    <div class="editor-actions">
+      <button class="button-secondary" @click="exportOpen = true">{{ t('导出', 'Export') }}</button>
      <span class="save-status" :class="editorStore.saveStatus">{{ statusText[editorStore.saveStatus] }}</span>
      <button v-if="needsRecovery && !missingFile" class="button-secondary" @click="reload">{{ t('重新加载外部版本', 'Reload external version') }}</button>
      <span v-if="missingFile" class="save-status conflict">{{ t('原文件已删除或移动', 'Original file deleted or moved') }}</span>
@@ -0,0 +1,56 @@
+<script setup lang="ts">
+import { ref, onMounted, onBeforeUnmount } from 'vue'
+import AppDialog from '@/components/common/AppDialog.vue'
+import { useEditorStore } from '@/stores/editor'
+import { useThemeStore } from '@/stores/theme'
+import { exportService, type ExportFormat, type ExportJob } from '@/services/exportService'
+const emit = defineEmits<{ close: [] }>()
+const editor = useEditorStore(), theme = useThemeStore()
+const format = ref<ExportFormat>('html'), page = ref('A4'), title = ref(true)
+const jobs = ref<ExportJob[]>([]), error = ref(''), preparing = ref(false)
+let timer: ReturnType<typeof setTimeout> | undefined, disposed = false
+let controller: AbortController | undefined
+const labels = { queued: '排队中', running: '渲染中', completed: '已完成', failed: '失败', cancelled: '已取消' }
+async function refresh() {
+  try { const value = await exportService.list(); if (!disposed) jobs.value = value } catch (e) { error.value = String(e) }
+  if (!disposed) timer = setTimeout(refresh, 1500)
+}
+async function start() {
+  preparing.value = true; error.value = ''; controller = new AbortController()
+  const snapshot = editor.content, name = editor.currentFilePath?.split('/').pop()?.replace(/\.md$/i, '') ?? '笔记'
+  try {
+    const job = await exportService.create(snapshot, name, format.value, { theme_id: theme.currentThemeId, include_title: title.value, page_size: page.value }, controller.signal, editor.currentFilePath ?? undefined)
+    if (!disposed) jobs.value.unshift(job)
+  } catch (e) { error.value = controller.signal.aborted ? '已取消图表准备' : String(e) }
+  finally { preparing.value = false }
+}
+async function action(job: ExportJob, download = false) {
+  try { if (download) await exportService.download(job); else await exportService.cancel(job.id) } catch (e) { error.value = String(e) }
+}
+onMounted(refresh)
+onBeforeUnmount(() => { disposed = true; clearTimeout(timer); controller?.abort() })
+</script>
+<template>
+  <AppDialog label="导出笔记" @close="emit('close')"><section class="modal export-modal">
+    <h2>导出笔记</h2><p>导出点击时的编辑器快照，包含未保存修改。关闭窗口后后台任务继续运行。</p>
+    <label for="export-format">格式</label><select id="export-format" v-model="format"><option value="html">HTML</option><option value="pdf">PDF</option><option value="docx">DOCX</option></select>
+    <label>纸张 <select v-model="page"><option>A4</option><option>Letter</option></select></label>
+    <label><input v-model="title" type="checkbox">包含标题</label>
+    <p v-if="format !== 'html'">PDF / DOCX 使用浅色打印样式。</p>
+    <button class="button-primary" :disabled="preparing || !editor.content.trim()" @click="start">{{ preparing ? '准备图表…' : '开始导出' }}</button>
+    <button v-if="preparing" @click="controller?.abort()">取消准备</button>
+    <p v-if="error" role="alert">{{ error }}</p>
+    <ul><li v-for="job in jobs" :key="job.id">
+      <strong>{{ job.fileName ?? job.id }}</strong> · {{ labels[job.status] }}
+      <button v-if="job.status === 'completed'" @click="action(job, true)">下载</button>
+      <button v-if="['queued','running'].includes(job.status)" @click="action(job)">取消</button>
+      <p v-if="job.error" role="alert">{{ job.error }}</p>
+      <ul v-if="job.warnings.length" class="export-warnings" aria-label="导出警告"><li v-for="warning in job.warnings" :key="warning">{{ warning }}</li></ul>
+    </li></ul>
+    <button class="button-secondary" @click="emit('close')">关闭</button>
+  </section></AppDialog>
+</template>
+<style scoped>
+.export-modal { width: min(640px, 100%); padding: 24px; background: var(--color-surface-primary); border: 1px solid var(--color-border-default); border-radius: 12px; }
+label { display: inline-flex; align-items:center; gap: 8px; margin: 8px; } li { margin-block: 12px; overflow-wrap: anywhere; } button { margin: 6px; } .export-warnings { color: var(--color-warning); } [role=alert] { color:var(--color-error); }
+</style>
@@ -156,20 +156,20 @@ function foldHeadings(action: 'toggle' | 'all' | 'none') {
    }
  })
 }
-const diagramPreviews = new Map<string, { source: string; apply: (value: HTMLElement) => void }>()
-function renderDiagram(source: string, apply: (value: HTMLElement) => void) {
+const diagramPreviews = new Map<string, { source: string; kind: string; apply: (value: HTMLElement) => void }>()
+function renderDiagram(source: string, apply: (value: HTMLElement) => void, kind = 'mermaid') {
  for (const [id, entry] of diagramPreviews) {
    if (entry.apply === apply) diagramPreviews.delete(id)
  }
-  const element = createMermaidPreview(source, themeStore.isDark, apply)
-  diagramPreviews.set(element.dataset.previewId!, { source, apply })
+  const element = createMermaidPreview(source, themeStore.isDark, apply, kind, themeStore.currentThemeId)
+  diagramPreviews.set(element.dataset.previewId!, { source, apply, kind })
  return element
 }
 watch(() => themeStore.currentThemeId, () => {
  const current = [...diagramPreviews.entries()]
  diagramPreviews.clear()
  for (const [id, entry] of current) {
-    if (editorRoot.value?.querySelector(`[id="${id}"]`)) entry.apply(renderDiagram(entry.source, entry.apply))
+    if (editorRoot.value?.querySelector(`[id="${id}"]`)) entry.apply(renderDiagram(entry.source, entry.apply, entry.kind))
  }
 }, { flush: 'post' })

@@ -333,8 +333,8 @@ onMounted(async () => {
    ...config,
    languages: shikiLanguages(themeStore.resolvedCodeBlockTheme),
    renderLanguage: renderCodeLanguage,
-    renderPreview: (language, content, applyPreview) => language.trim().toLowerCase() === 'mermaid'
-      ? markdownPreferences.diagrams ? renderDiagram(content, applyPreview) : null
+    renderPreview: (language, content, applyPreview) => ['mermaid', 'function-plot'].includes(language.trim().toLowerCase())
+      ? markdownPreferences.diagrams ? renderDiagram(content, applyPreview, language.trim().toLowerCase()) : null
      : config.renderPreview(language, content, applyPreview),
    extensions: [basicSetup, keymap.of([indentWithTab]), shikiEditorTheme(themeStore.resolvedCodeBlockTheme),
      indentUnit.of(' '.repeat(markdownPreferences.indent)), CodeEditorState.tabSize.of(markdownPreferences.indent),
@@ -1,17 +1,18 @@
+import { renderFunctionPlot } from '@/services/functionPlotService'
 import { nextTick } from 'vue'
 import { renderMermaid } from '@/services/mermaidService'
 import { t } from '@/i18n'
 import { appendDiagramControls } from '@/utils/diagramControls'

 let previewId = 0
-export function createMermaidPreview(source: string, dark: boolean, applyPreview: (value: HTMLElement) => void): HTMLElement {
+export function createMermaidPreview(source: string, dark: boolean, applyPreview: (value: HTMLElement) => void, kind = 'mermaid', themeId = 'light'): HTMLElement {
  // Each revision owns its element, so a slow render cannot replace newer content.
  // Milkdown sanitizes Element input to its inner HTML; retain the revision
  // marker and controls inside an otherwise disposable envelope.
  const envelope = document.createElement('div')
  const container = document.createElement('div')
  envelope.append(container)
-  container.className = 'editor-mermaid-preview'
+  container.className = 'editor-mermaid-preview' + (kind === 'function-plot' ? ' function-plot-preview' : '')
  container.id = `editor-mermaid-preview-${++previewId}`
  envelope.dataset.previewId = container.id
  container.setAttribute('aria-live', 'polite')
@@ -27,8 +28,11 @@ export function createMermaidPreview(source: string, dark: boolean, applyPreview
      applyPreview(envelope.cloneNode(true) as HTMLElement)
    }
  }
-  void renderMermaid(source, { theme: dark ? 'dark' : 'light' }).then(result => {
-    if (result.warnings.length) {
+  void (kind === 'function-plot' ? new Promise<void>(resolve => setTimeout(resolve, 180)) : Promise.resolve()).then<{ svg: string; warnings: string[] }>(() => {
+    if (kind === 'function-plot' && !document.getElementById(container.id)) throw new Error('stale preview')
+    return kind === 'function-plot' ? renderFunctionPlot(source, themeId) : renderMermaid(source, { theme: dark ? 'dark' : 'light' })
+  }).then(result => {
+    if (result.warnings.length && (kind === 'mermaid' || !result.svg)) {
      container.classList.add('has-error')
      container.textContent = `${t('图表语法有误，可点击编辑修改：', 'Diagram syntax error. Choose Edit to fix:')} ${result.warnings.join('\n')}`
      void publish()
@@ -37,6 +41,7 @@ export function createMermaidPreview(source: string, dark: boolean, applyPreview
    // Mermaid runs in strict mode; Milkdown sanitizes the preview before insertion.
    container.innerHTML = result.svg
    appendDiagramControls(container)
+    if (result.warnings.length) { const warning = document.createElement('p'); warning.textContent = result.warnings.join('\n'); warning.setAttribute('role', 'status'); container.append(warning) }
    void publish()
  }).catch(() => {
    container.textContent = t('图表渲染失败，请点击编辑检查源码。', 'Unable to render diagram. Choose Edit to inspect the source.')
@@ -69,7 +69,7 @@ it('offers fenced-code aliases and retains the LaTeX selector', () => {

 it('offers every bundled Shiki language and alias', () => {
  const languages = shikiLanguages('github-light')
-  expect(languages).toHaveLength(bundledLanguagesInfo.length + 1)
+  expect(languages).toHaveLength(bundledLanguagesInfo.length + 2)
  for (const info of bundledLanguagesInfo) {
    const language = languages.find(item => item.alias.includes(info.id))!
    expect(language, info.id).toBeDefined()
@@ -60,6 +60,7 @@ export async function shikiLanguage(language: string, theme: CodeTheme): Promise

 export function shikiLanguages(theme: CodeTheme): LanguageDescription[] {
  return [
+    LanguageDescription.of({ name: 'function-plot', alias: ['Function Plot'], load: () => shikiLanguage('text', theme) }),
    ...bundledLanguagesInfo.map(info => LanguageDescription.of({
      name: info.id,
      alias: [info.name, ...(info.aliases ?? [])],
@@ -3,6 +3,7 @@ import { useWorkspaceStore } from '@/stores/workspace'
 import { t } from '@/i18n'

 const routes = [
+  { path: '/benchmarks', name: 'benchmarks', component: () => import('@/features/benchmarks/BenchmarkView.vue'), meta: { title: 'Benchmark' } },
  { path: '/logs', name: 'logs', component: () => import('@/features/logs/LogsView.vue'), meta: { title: '运行日志' } },
  { path: '/media', name: 'media', component: () => import('@/features/media/MediaView.vue'), meta: { title: '音视频转写', requiresVault: true } },
  {
@@ -80,7 +81,10 @@ const router = createRouter({

 router.beforeEach((to) => {
  const workspaceStore = useWorkspaceStore()
-  if (to.meta.requiresVault && !workspaceStore.hasVault) {
+  // A benchmark can run without the workspace UI being open. Its persisted Trace
+  // and permission tickets must remain reachable from the report page.
+  const existingAgentRun = to.name === 'agent' && Boolean(to.params.runId)
+  if (to.meta.requiresVault && !workspaceStore.hasVault && !existingAgentRun) {
    return { path: '/' }
  }
  if (to.path === '/' && workspaceStore.hasVault) {
@@ -0,0 +1,14 @@
+import { apiClient } from './apiClient'
+interface RunWire { run_id: string; kind: 'rag' | 'agent'; dataset_id: string; status: string; progress: number | null; config_snapshot: Record<string, unknown>; error_code: string | null }
+export interface BenchmarkRun { id: string; kind: 'rag' | 'agent'; datasetId: string; status: string; progress: number | null; agentId?: string; errorCode: string | null }
+const map = (r: RunWire): BenchmarkRun => ({ id: r.run_id, kind: r.kind, datasetId: r.dataset_id, status: r.status, progress: r.progress, agentId: r.config_snapshot.active_agent_run_id as string | undefined, errorCode: r.error_code })
+export const benchmarkService = {
+  async datasets(kind: 'rag' | 'agent') {
+    const r = await apiClient.get<{ items: { dataset_id: string; description: string; case_count: number }[] }>('/api/benchmarks/datasets', { params: { kind } })
+    return r.items.map(d => ({ id: d.dataset_id, description: d.description, cases: d.case_count }))
+  },
+  async list() { return (await apiClient.get<{ items: RunWire[] }>('/api/benchmarks/runs')).items.map(map) },
+  async start(kind: 'rag' | 'agent', body: object) { return map(await apiClient.post<RunWire>(`/api/benchmarks/${kind}/runs`, body)) },
+  cancel(id: string) { return apiClient.post(`/api/benchmarks/runs/${encodeURIComponent(id)}/cancel`) },
+  report(id: string) { return apiClient.get<{ metrics: Record<string, unknown>; cases: unknown[]; config_snapshot: Record<string, unknown> }>(`/api/benchmarks/runs/${encodeURIComponent(id)}/report`) },
+}
@@ -0,0 +1,18 @@
+import {describe,it,expect,vi} from 'vitest'
+vi.mock('./apiClient',()=>({apiClient:{post:vi.fn(),get:vi.fn()}}))
+import {apiClient} from './apiClient'
+import {exportService} from './exportService'
+describe('export snapshot contract',()=>{
+  it('submits the unsaved Markdown snapshot and maps warnings and filename',async()=>{
+    vi.mocked(apiClient.post).mockResolvedValue({job_id:'job',status:'queued',warnings:['print palette'],file:{file_name:'note.pdf'},error:null})
+    const job=await exportService.create('# unsaved', 'note','pdf',{theme_id:'dark',include_title:true,page_size:'A4'},undefined,'folder/note.md')
+    expect(apiClient.post).toHaveBeenCalledWith('/api/exports',expect.objectContaining({source:{type:'markdown',markdown:'# unsaved',file_path:'folder/note.md'},format:'pdf'}))
+    expect(job.warnings).toEqual(['print palette']);expect(job.fileName).toBe('note.pdf')
+  })
+  it('aborted preparation does not create a backend job',async()=>{
+    vi.mocked(apiClient.post).mockClear()
+    const abort=new AbortController();abort.abort()
+    await expect(exportService.create('snapshot','note','html',{theme_id:'light',include_title:true,page_size:'A4'},abort.signal)).rejects.toThrow()
+    expect(apiClient.post).not.toHaveBeenCalled()
+  })
+})
@@ -0,0 +1,68 @@
+import { apiClient } from './apiClient'
+import { Marked } from 'marked'
+import { renderMermaid } from './mermaidService'
+
+export type ExportFormat = 'html' | 'pdf' | 'docx'
+interface JobWire {
+  job_id: string; status: 'queued' | 'running' | 'completed' | 'failed' | 'cancelled'
+  warnings: string[]; error: string | null
+  file: { file_name: string; size: number } | null
+}
+export interface ExportJob { id: string; status: JobWire['status']; warnings: string[]; error: string | null; fileName?: string }
+const mapJob = (w: JobWire): ExportJob => ({ id: w.job_id, status: w.status, warnings: w.warnings, error: w.error, fileName: w.file?.file_name })
+export async function rasterize(svg: string, signal?: AbortSignal): Promise<string> {
+  const doc = new DOMParser().parseFromString(svg, 'image/svg+xml')
+  const root = doc.documentElement
+  const box = root.getAttribute('viewBox')?.split(/[ ,]+/).map(Number)
+  const width = box?.[2] || 800, height = box?.[3] || 600
+  if (!Number.isFinite(width + height) || width <= 0 || height <= 0) throw new Error('图表尺寸无效')
+  const scale = Math.min(4, Math.max(2, 1200 / width), Math.sqrt(4_000_000 / (width * height)))
+  root.setAttribute('width', String(Math.floor(width * scale))); root.setAttribute('height', String(Math.floor(height * scale)))
+  root.style.maxWidth = 'none'
+  const data = new XMLSerializer().serializeToString(root)
+  const image = new Image()
+  image.src = `data:image/svg+xml;charset=utf-8,${encodeURIComponent(data)}`
+  await new Promise<void>((resolve, reject) => {
+    const abort = () => reject(new DOMException('Aborted', 'AbortError'))
+    const timer = setTimeout(() => reject(new Error('图表图片解码超时')), 15000)
+    const cleanup = () => { clearTimeout(timer); signal?.removeEventListener('abort', abort) }
+    if (signal?.aborted) { cleanup(); abort(); return }
+    signal?.addEventListener('abort', abort, { once: true })
+    image.decode().then(resolve, reject).finally(cleanup)
+  })
+  const canvas = document.createElement('canvas')
+  canvas.width = Math.floor(width * scale); canvas.height = Math.floor(height * scale)
+  const context = canvas.getContext('2d')!
+  context.fillStyle = '#ffffff'; context.fillRect(0, 0, canvas.width, canvas.height)
+  context.drawImage(image, 0, 0, canvas.width, canvas.height)
+  return canvas.toDataURL('image/png').split(',')[1]!
+}
+export async function hashSource(source: string) {
+  return [...new Uint8Array(await crypto.subtle.digest('SHA-256', new TextEncoder().encode(source.trim())))].map(v => v.toString(16).padStart(2, '0')).join('')
+}
+export const exportService = {
+  async create(markdown: string, title: string, format: ExportFormat, options: { theme_id: string; include_title: boolean; page_size: string }, signal?: AbortSignal, filePath?: string) {
+    const blocks: string[] = []
+    const parser = new Marked()
+    parser.walkTokens(parser.lexer(markdown), token => { if (token.type === 'code' && token.lang === 'mermaid') blocks.push(token.text) })
+    const assets = []
+    for (const source of [...new Set(blocks)]) {
+      signal?.throwIfAborted()
+      if (assets.length >= 16) throw new Error('每次导出最多 16 个 Mermaid 图表')
+      const result = await renderMermaid(source, { mode: 'raster', theme: 'light' })
+      if (result.warnings.length) throw new Error(`Mermaid 无法导出：${result.warnings.join('; ')}`)
+      assets.push({ kind: 'mermaid', source_hash: await hashSource(source), png_base64: await rasterize(result.svg, signal) })
+    }
+    signal?.throwIfAborted()
+    return mapJob(await apiClient.post<JobWire>('/api/exports', { source: { type: 'markdown', markdown, file_path: filePath }, title, format, options, assets }))
+  },
+  async get(id: string) { return mapJob(await apiClient.get<JobWire>(`/api/exports/${encodeURIComponent(id)}`)) },
+  async list() { const response = await apiClient.get<{ items: JobWire[] }>('/api/exports'); return response.items.map(mapJob) },
+  cancel(id: string) { return apiClient.post(`/api/exports/${encodeURIComponent(id)}/cancel`) },
+  async download(job: ExportJob) {
+    const response = await apiClient.get<Response>(`/api/exports/${encodeURIComponent(job.id)}/file`)
+    const url = URL.createObjectURL(await response.blob())
+    const link = document.createElement('a'); link.href = url; link.download = job.fileName ?? 'export'
+    link.click(); setTimeout(() => URL.revokeObjectURL(url), 1000)
+  },
+}
@@ -0,0 +1,22 @@
+// @vitest-environment jsdom
+import { describe,it,expect,vi,beforeEach } from 'vitest'
+vi.mock('./apiClient',()=>({apiClient:{post:vi.fn()}}))
+import { apiClient } from './apiClient'
+import { renderFunctionPlot } from './functionPlotService'
+describe('function plot preview boundary',()=>{
+  beforeEach(()=>vi.clearAllMocks())
+  it('shares requests only for the same source and theme, sanitizes SVG',async()=>{
+    vi.mocked(apiClient.post).mockResolvedValue({result:{content:'<svg onload="alert(1)"><script>alert(2)</script><path d="M0 0L1 1"/></svg>',warnings:[]},diagnostics:[],node_count:3})
+    const a=await renderFunctionPlot('y = x + 100','dark')
+    await renderFunctionPlot('y = x + 100','dark')
+    await renderFunctionPlot('y = x + 100','light')
+    expect(apiClient.post).toHaveBeenCalledTimes(2)
+    expect(a.svg).not.toMatch(/onload|script|alert/)
+  })
+  it('does not cache network failures',async()=>{
+    vi.mocked(apiClient.post).mockRejectedValueOnce(new Error('offline')).mockResolvedValueOnce({result:null,diagnostics:[{message:'bad expression',line:2}],node_count:0})
+    await expect(renderFunctionPlot('bad expression')).rejects.toThrow('offline')
+    const result=await renderFunctionPlot('bad expression')
+    expect(result.svg).toBe('');expect(result.warnings[0]).toContain('2')
+  })
+})
@@ -0,0 +1,21 @@
+import { apiClient } from './apiClient'
+import DOMPurify from 'dompurify'
+
+interface PlotWire {
+  result: { content: string; width: number; height: number; warnings: string[] } | null
+  diagnostics: { message: string; severity: string; line?: number }[]
+  node_count: number
+}
+const cache = new Map<string, Promise<{ svg: string; warnings: string[]; nodeCount: number }>>()
+export function renderFunctionPlot(source: string, themeId = 'light') {
+  const key = JSON.stringify([source, themeId])
+  if (cache.has(key)) return cache.get(key)!
+  const result = apiClient.post<PlotWire>('/api/plots/function', { source, theme_id: themeId }, { timeoutMs: 30000 }).then(wire => ({
+    svg: DOMPurify.sanitize(wire.result?.content ?? '', { USE_PROFILES: { svg: true } }),
+    warnings: [...wire.diagnostics.map(d => `${d.message}${d.line ? ` (行 ${d.line})` : ''}`), ...(wire.result?.warnings ?? [])],
+    nodeCount: wire.node_count,
+  })).catch(error => { cache.delete(key); throw error })
+  if (cache.size >= 32) cache.delete(cache.keys().next().value!)
+  cache.set(key, result)
+  return result
+}
@@ -8,8 +8,8 @@ function loadMermaid() {
 import { computed } from 'vue'
 import { useThemeStore } from '@/stores/theme'

-export function mermaidThemeVariables(dark: boolean) {
-  const style = typeof document === 'undefined' ? null : getComputedStyle(document.documentElement)
+export function mermaidThemeVariables(dark: boolean, useDocument = true) {
+  const style = !useDocument || typeof document === 'undefined' ? null : getComputedStyle(document.documentElement)
  const color = (name: string, fallback: string) => style?.getPropertyValue(`--color-${name}`).trim() || fallback
  const text = color('text-primary', dark ? '#e6edf3' : '#1f2328')
  const border = color('border-default', dark ? '#484f58' : '#d0d7de')
@@ -29,15 +29,15 @@ export function mermaidThemeVariables(dark: boolean) {
  }
 }

-async function ensureInitialized(theme: 'light' | 'dark') {
+async function ensureInitialized(theme: 'light' | 'dark', raster = false) {
    const mermaid = await loadMermaid()
    mermaid.initialize({
      startOnLoad: false,
      theme: 'base',
-      themeVariables: mermaidThemeVariables(theme === 'dark'),
+      themeVariables: mermaidThemeVariables(theme === 'dark', !raster),
      securityLevel: 'strict',
-      fontFamily: 'var(--font-ui-sans)',
-      flowchart: { useMaxWidth: true, htmlLabels: true },
+      fontFamily: raster ? 'Arial, Microsoft YaHei, sans-serif' : 'var(--font-ui-sans)',
+      flowchart: { useMaxWidth: true, htmlLabels: !raster },
      sequence: { useMaxWidth: true },
      gantt: { useMaxWidth: true },
    })
@@ -65,18 +65,18 @@ export interface MermaidParseError {

 let renderCounter = 0

-export function renderMermaid(source: string, options: { theme?: 'light' | 'dark'; mode?: 'interactive' | 'static' } = {}): Promise<MermaidRenderResult> {
+export function renderMermaid(source: string, options: { theme?: 'light' | 'dark'; mode?: 'interactive' | 'static' | 'raster' } = {}): Promise<MermaidRenderResult> {
  return serialized(() => renderMermaidNow(source, options))
 }

 async function renderMermaidNow(
  source: string,
-  options: { theme?: 'light' | 'dark'; mode?: 'interactive' | 'static' } = {}
+  options: { theme?: 'light' | 'dark'; mode?: 'interactive' | 'static' | 'raster' } = {}
 ): Promise<MermaidRenderResult> {
  const theme = options.theme ?? 'light'
  const id = `mermaid-${Date.now()}-${++renderCounter}`
  try {
-    const mermaid = await ensureInitialized(theme)
+    const mermaid = await ensureInitialized(theme, options.mode === 'raster')
    const result = await mermaid.render(id, source)
    const parser = new DOMParser()
    const doc = parser.parseFromString(result.svg, 'image/svg+xml')
@@ -375,3 +375,41 @@ ol {
  --color-border-focus: #8a5b32;
  --color-border-disabled: #eadfc4;
 }
+
+/* Function plot tokens inherit all installed themes, including custom packages. */
+:root {
+  --color-plot-background: var(--color-surface-primary);
+  --color-plot-text: var(--color-text-primary);
+  --color-plot-axis: var(--color-text-secondary);
+  --color-plot-grid: var(--color-border-default);
+  --color-plot-curve-0: #0969da;
+  --color-plot-curve-1: #d1242f;
+  --color-plot-curve-2: #1a7f37;
+  --color-plot-curve-3: #8250df;
+  --color-plot-curve-4: #9a6700;
+  --color-plot-curve-5: #bc4c00;
+}
+[data-theme='dark'], [data-theme='midnight-purple'] {
+  --color-plot-curve-0: #79c0ff;
+  --color-plot-curve-1: #ff9b9b;
+  --color-plot-curve-2: #7ee787;
+  --color-plot-curve-3: #d2a8ff;
+  --color-plot-curve-4: #f2cc60;
+  --color-plot-curve-5: #ffa657;
+}
+.function-plot-svg > rect:first-child { fill: var(--color-plot-background); }
+.function-plot-svg .plot-grid { stroke: var(--color-plot-grid); }
+.function-plot-svg .plot-axis { stroke: var(--color-plot-axis); }
+.function-plot-svg text { fill: var(--color-plot-text); }
+.function-plot-svg .plot-curve-0 { stroke: var(--color-plot-curve-0); }
+.function-plot-svg .plot-legend-0 { fill: var(--color-plot-curve-0); }
+.function-plot-svg .plot-curve-1 { stroke: var(--color-plot-curve-1); }
+.function-plot-svg .plot-legend-1 { fill: var(--color-plot-curve-1); }
+.function-plot-svg .plot-curve-2 { stroke: var(--color-plot-curve-2); }
+.function-plot-svg .plot-legend-2 { fill: var(--color-plot-curve-2); }
+.function-plot-svg .plot-curve-3 { stroke: var(--color-plot-curve-3); }
+.function-plot-svg .plot-legend-3 { fill: var(--color-plot-curve-3); }
+.function-plot-svg .plot-curve-4 { stroke: var(--color-plot-curve-4); }
+.function-plot-svg .plot-legend-4 { fill: var(--color-plot-curve-4); }
+.function-plot-svg .plot-curve-5 { stroke: var(--color-plot-curve-5); }
+.function-plot-svg .plot-legend-5 { fill: var(--color-plot-curve-5); }
@@ -1,3 +1,4 @@
+import { renderFunctionPlot } from '@/services/functionPlotService'
 import DOMPurify from 'dompurify'
 import { Marked } from 'marked'
 import { defaultMarkdownPreferences, type MarkdownPreferences } from '@/stores/markdownPreferences'
@@ -125,7 +126,7 @@ export async function getCodeTokenizer(theme: 'github-light' | 'github-dark', re
  }
 }

-export async function renderMarkdown(source: string, options?: { theme?: 'light' | 'dark'; preferences?: MarkdownPreferences; citationNumbers?: number[]; citationAliases?: Record<string, number> }): Promise<string> {
+export async function renderMarkdown(source: string, options?: { themeId?: string; theme?: 'light' | 'dark'; preferences?: MarkdownPreferences; citationNumbers?: number[]; citationAliases?: Record<string, number> }): Promise<string> {
  const preferences = options?.preferences ?? defaultMarkdownPreferences
  const marked = createMarkdownParser(preferences)
  const citations = new Set(options?.citationNumbers ?? [])
@@ -141,12 +142,12 @@ export async function renderMarkdown(source: string, options?: { theme?: 'light'
  const html = marked.parse(source, { async: false }) as string
  const documentNode = new DOMParser().parseFromString(`<body>${html}</body>`, 'text/html')

-  const mermaidBlocks: { pre: Element; source: string }[] = []
+  const mermaidBlocks: { pre: Element; source: string; kind: string }[] = []

  for (const code of documentNode.querySelectorAll('pre > code')) {
    const requestedLanguage = [...code.classList].find((name) => name.startsWith('language-'))?.slice(9) || 'text'
-    if (requestedLanguage === 'mermaid' && preferences.diagrams) {
-      mermaidBlocks.push({ pre: code.parentElement!, source: code.textContent ?? '' })
+    if (['mermaid', 'function-plot'].includes(requestedLanguage) && preferences.diagrams) {
+      mermaidBlocks.push({ pre: code.parentElement!, source: code.textContent ?? '', kind: requestedLanguage })
      continue
    }
    if (requestedLanguage.toLowerCase() === 'latex' && preferences.math) {
@@ -168,19 +169,23 @@ export async function renderMarkdown(source: string, options?: { theme?: 'light'
    code.parentElement?.replaceWith(wrapper)
  }

-  for (const { pre, source } of mermaidBlocks) {
+  let plotCount = 0, plotNodes = 0
+  for (const { pre, source, kind } of mermaidBlocks) {
    try {
-      const result = await renderMermaid(source, { theme: options?.theme, mode: 'static' })
+      if (kind === 'function-plot' && ++plotCount > 16) throw new Error('函数图像数量超过 16')
+      const result = kind === 'function-plot' ? await renderFunctionPlot(source, options?.themeId) : await renderMermaid(source, { theme: options?.theme, mode: 'static' })
+      if ('nodeCount' in result && (plotNodes += result.nodeCount) > 8000) throw new Error('函数图像累计复杂度超过 8000')
      const container = document.createElement('div')
-      container.className = 'markdown-mermaid'
+      container.className = 'markdown-mermaid' + (kind === 'function-plot' ? ' markdown-function-plot' : '')
      container.innerHTML = result.svg
-      appendCodeToolbar(container, 'mermaid', source, true)
-      if (!result.warnings.length) appendDiagramControls(container)
+      appendCodeToolbar(container, kind, source, true)
+      if (result.warnings.length) { const message = document.createElement('p'); message.textContent = result.warnings.join('\n'); message.setAttribute('role', 'status'); container.append(message) }
+      if (!result.warnings.length || (kind === 'function-plot' && result.svg)) appendDiagramControls(container)
      pre.replaceWith(container)
-    } catch {
+    } catch (error) {
      const fallback = document.createElement('pre')
      fallback.className = 'mermaid-error'
-      fallback.textContent = source
+      fallback.textContent = `${error instanceof Error ? error.message : '图表渲染失败'}\n${source}`
      pre.replaceWith(fallback)
    }
  }