AgentRagPipeline

Five-stage retrieval-augmented generation: progress, source cards, and citations that point back to sources.

Demo

AgentRagPipelineOne public API for H5 and Wevu

Live state

Release readiness workflowAgentRagPipeline

检索增强生成

流程已完成

2.9 s

1. 理解问题380 ms已完成

   确认本次查询与来源范围

   How can we make signup failures recoverable?

2. 向量编码650 ms已完成

   将问题转换为检索表示
3. 检索来源540 ms已完成

   查找与问题相关的片段
4. 组装上下文580 ms已完成

   将来源片段组织为上下文

   1. 1来源 1：客服工单
   2. 2来源 2：会话回放
5. 引用回答720 ms已完成

   生成可追溯到来源的回答

   验证链接过期后应提供新链接。 会话回放也显示缺少恢复入口。

Basic Usage

<script setup lang="ts">
import { AgentRagPipeline } from '@/components/agent-ui'
</script>

<template>
  <AgentRagPipeline
    :query="query"
    :steps="steps"
    :sources="sources"
    :answer="answer"
    @run="run"
    @cancel="cancel"
    @select-source="selectSource"
  />
</template>

Props

Prop	Type	Default	Description
className	ClassValue	undefined	Root classes
query	string	''	Current query
steps	readonly AgentRagStep[]	[]	Five-stage progress
sources	readonly AgentRagSource[]	[]	Retrieved sources
answer	readonly AgentRagAnswerPart[]	[]	Text fragments and source citations
title	string	'检索增强生成'	Title
elapsedMs	number	undefined	Elapsed time
reducedMotion	boolean	false	Disable non-essential motion

Events

Event	Payload	Description
run	void	Start or replay
cancel	void	Stop the current run
selectSource	AgentRagSource	Select a source

Target notes

Target	Import
H5	Named export from @/components/agent-ui
weapp	Default export from @/components/agent-ui/AgentRagPipeline.vue

The docs demo uses a local snapshot and does not call a model or retrieval service.