feat: 支持浏览器内置大模型

packages/doc-ai/package.json

@@ -30,7 +30,7 @@
   "devDependencies": {
     "@types/node": "^22.15.16",
     "@vitejs/plugin-vue": "^5.2.2",
-    "@vitejs/plugin-vue-jsx": "^5.0.1",
+    "@vitejs/plugin-vue-jsx": "^5.1.1",
     "@vue/tsconfig": "^0.7.0",
     "@vant/auto-import-resolver": "^1.3.0",
     "dotenv": "^16.5.0",

packages/doc-ai/src/const.ts

@@ -1,3 +1,3 @@
-export const AGENT_ROOT = 'https://agent.opentiny.design/api/v1/webmcp-trial/'
+export const AGENT_ROOT = 'https://ai.opentiny.design/'
 
-export const SSEION_ID = 'spgl-95c0-4839-8007-remoter'
+export const SSEION_ID = 'f5d8e6f6-8f9a-4117-ad9c-b9f10b98068e'

packages/doc-ai/src/views/comprehensive/index.vue

@@ -101,6 +101,7 @@ server.registerTool(
     }
   },
   async ({ city }: { city: string }) => {
+    debugger
     return {
       content: [{ type: 'text', text: `天气信息：${city}晴天` }]
     }

packages/next-remoter/package.json

@@ -25,6 +25,7 @@
     "@modelcontextprotocol/sdk": "^1.16.0",
     "@opentiny/vue": "^3.25.0",
     "@opentiny/vue-icon": "^3.25.0",
+    "@built-in-ai/core": "^2.0.0",
     "html5-qrcode": "^2.3.8",
     "vant": "^4.9.21",
     "vue": "^3.3.11"

packages/next-remoter/src/components/tiny-robot-chat.vue

@@ -125,7 +125,7 @@ const props = defineProps({
   /** 后端的代理服务器地址 */
   agentRoot: {
     type: String,
-    default: 'https://agent.opentiny.design/api/v1/webmcp-trial/'
+    default: 'https://ai.opentiny.design/'
   },
   /** 左上角的标题 */
   title: {
@@ -229,7 +229,7 @@ const pillItems = [
     id: 'office',
     text: props.locale === 'zh-CN' ? '办公助手' : 'Office Assistant',
     menus: [
-      '接收邮件#请同步邮箱的新邮件。',
+      '接收邮件#帮我勾选中最贵的手机商品。',
       '编写邮件#请新建一个邮件，收件人为 opentiny-next@meeting.com, 内容为举办一个临时会议。',
       '安排会议#创建一个临时的在线会议，主题为讨论问题，时长为1小时。',
       '整理文档#请分析附件中的销售情况，把销售额绘制成折线图。'

packages/next-remoter/src/composable/CustomAgentModelProvider.ts

@@ -5,6 +5,7 @@ import { BaseModelProvider } from '@opentiny/tiny-robot-kit'
 import type { AIModelConfig } from '@opentiny/tiny-robot-kit'
 import { type Ref } from 'vue'
 import { AgentModelProvider, McpServerConfig, IAgentModelProviderOption } from '@opentiny/next-sdk'
+import { builtInAI } from '@built-in-ai/core';
 
 /** Tiny-robot 所需要的自定义大语言的Provider */
 export class CustomAgentModelProvider extends BaseModelProvider {
@@ -14,11 +15,8 @@ export class CustomAgentModelProvider extends BaseModelProvider {
   constructor(config: AIModelConfig, sessionId: Ref<string>, agentRoot: Ref<string>) {
     super(config)
     const options = {
-      llmConfig: {
-        apiKey: 'sk-trial',
-        baseURL: 'https://agent.opentiny.design/api/v1/ai',
-        providerType: 'deepseek'
-      },
+      llm: builtInAI,
+      isReActModel: true,
       mcpServers: [] as McpServerConfig[]
     }
     if (sessionId.value && sessionId.value.includes(',')) {
@@ -46,14 +44,15 @@ export class CustomAgentModelProvider extends BaseModelProvider {
   async chatStream(request: ChatCompletionRequest, handler: StreamHandler): Promise<void> {
     const result = await this.agent.chatStream({
       messages: request.messages,
-      model: 'deepseek-ai/DeepSeek-V3',
-      abortSignal: request.options?.signal
+      model: builtInAI(),
+      abortSignal: request.options?.signal,
+      handler
     })
 
     // 标识每一个markdown块
     let textId = 1
     for await (const part of result.fullStream) {
-      // console.log(part, part.type)
+      console.log(part, part.type)
 
       // 文本节点处理。 每个文本块拥有自己的textId
       if (part.type === 'text-start') {

packages/next-remoter/src/composable/useTinyRobotChat.ts

@@ -29,7 +29,7 @@ export const useTinyRobotChat = ({ sessionId, agentRoot }: useTinyRobotOption) =
     events: {
       onReceiveData(data, messages, preventDefault) {
         preventDefault()
-        console.log('onReceiveData=', data)
+        // console.log('onReceiveData=', data)
 
         let lastMessage = messages.value[messages.value.length - 1]
 

packages/next-sdk/README.md

@@ -11,3 +11,70 @@ npm install @opentiny/next-sdk
 ## 许可证
 
 MIT
+
+// onChunk: (chunkObj) => {
+// console.log("onChunk", chunkObj);
+// },
+// // onFinish: (finishObj) => console.log("onFinish", finishObj),
+// onStepFinish: (stepFinishObj) => console.log("onStepFinish", stepFinishObj),
+// experimental_transform:
+// //@ts-ignore
+// ({ stopStream }) => {
+// return new TransformStream({
+// transform(chunk, controller) {
+// console.log("transform chunk", chunk, controller);
+// if (
+// chunk.type == "tool-result" &&
+// chunk.toolName == "getOrderbyDate" &&
+// chunk.result.includes("没有")
+// ) {
+// // 停止流
+// stopStream();
+// console.log("已经停止流",chunk, result)
+// setTimeout(() => {
+// let askResult=confirm(chunk.result+", 你是否继续预订其它日期？")
+// if(askResult){
+// }
+// }, 100);
+// //@ts-ignore
+// controller.enqueue({
+// type: "step-finish",
+// finishReason: "stop",
+// logprobs: undefined,
+// usage: {
+// completionTokens: NaN,
+// promptTokens: NaN,
+// totalTokens: NaN,
+// },
+// request: {},
+// response: {
+// id: "response-id",
+// modelId: "mock-model-id",
+// timestamp: new Date(0),
+// },
+// warnings: [],
+// isContinued: false,
+// });
+// //@ts-ignore
+// controller.enqueue({
+// type: "finish",
+// finishReason: "stop",
+// logprobs: undefined,
+// usage: {
+// completionTokens: NaN,
+// promptTokens: NaN,
+// totalTokens: NaN,
+// },
+// response: {
+// id: "response-id",
+// modelId: "mock-model-id",
+// timestamp: new Date(0),
+// },
+// });
+// return;
+// }
+// controller?.enqueue(chunk);
+// },
+// });
+// },
+// });

packages/next-sdk/agent/AgentModelProvider.ts

@@ -7,6 +7,7 @@ import { ProviderV2 } from '@ai-sdk/provider'
 import { OpenAIProvider } from '@ai-sdk/openai'
 import { createOpenAI } from '@ai-sdk/openai'
 import { createDeepSeek } from '@ai-sdk/deepseek'
+import { getSystemPromptMessages, organizeToolCalls, runReActLoop } from './react'
 
 export const AIProviderFactories = {
   ['openai']: createOpenAI,
@@ -30,11 +31,15 @@ export class AgentModelProvider {
   mcpTools: Array<Record<string, any>> = []
   /**  需要实时过滤掉的tools name*/
   ignoreToolnames: string[] = []
+  /** 是否是 ReAct 模型 */
+  isReActModel: boolean = false
 
-  constructor({ llmConfig, mcpServers, llm }: IAgentModelProviderOption) {
+  constructor({ llmConfig, mcpServers, llm, isReActModel }: IAgentModelProviderOption) {
     // 1、保存 mcpServer
     this.mcpServers = mcpServers || []
 
+    this.isReActModel = isReActModel || false
+
     // 2、保存 llm
     if (llm) {
       this.llm = llm
@@ -156,11 +161,28 @@ export class AgentModelProvider {
       throw new Error('LLM is not initialized')
     }
 
+    const tools = (await this.tempMergeTools(options.tools)) as ToolSet
+    const systemPrompt = await getSystemPromptMessages(tools)
+    const llm = this.llm(model)
     return chatMethod({
       // @ts-ignore  ProviderV2 是所有llm的父类， 在每一个具体的llm 类都有一个选择model的函数用法
-      model: this.llm(model),
-      tools: this.tempMergeTools(options.tools) as ToolSet,
+      model: llm,
+      system: systemPrompt,
+      tools: this.isReActModel ? (tools as ToolSet) : undefined,
       stopWhen: stepCountIs(maxSteps),
+      onStepFinish: async (step) => {
+        if (this.isReActModel) {
+          await runReActLoop({
+            step,
+            tools,
+            vm: this,
+            chatMethod,
+            llm,
+            options,
+            system: systemPrompt
+          })
+        }
+      },
       ...options
     })
   }

packages/next-sdk/agent/react/index.ts

@@ -0,0 +1,149 @@
+import { Tool, ToolCall } from '../type'
+import { PREFIX, FORMAT_INSTRUCTIONS, SUFFIX } from './systemPrompt'
+import { generateText, stepCountIs } from 'ai'
+
+const FINAL_ANSWER_TAG = 'Final Answer:'
+const ACTION_TAG = '"action":'
+
+export const getSystemPromptMessages = async (tools: Tool[]): Promise<string> => {
+  const toolStrings = JSON.stringify(tools)
+  const prompt = [PREFIX, toolStrings, FORMAT_INSTRUCTIONS, SUFFIX].join('\n\n')
+
+  return prompt
+}
+
+export const organizeToolCalls = async (
+  text: string
+): Promise<{ toolCalls: ToolCall[]; thought?: string; finalAnswer: string }> => {
+  try {
+    let thought: string | undefined
+    const thoughtActionRegex = /Thought(.*?)(?:Action|Final Answer|$)/gs
+    const matches = [...text.matchAll(thoughtActionRegex)]
+
+    if (matches.length > 0) {
+      // 取第一个 Thought 作为思考内容，去除首尾的符号
+      thought = matches[0][1]?.replace(/^\W|$/, '')?.trim()
+    }
+
+    if (text.includes(FINAL_ANSWER_TAG) && !text.includes(ACTION_TAG)) {
+      const parts = text.split(FINAL_ANSWER_TAG)
+      const output = parts[parts.length - 1].trim()
+      return {
+        toolCalls: [],
+        thought,
+        finalAnswer: output
+      }
+    }
+
+    if (!text.includes(ACTION_TAG)) {
+      return {
+        toolCalls: [],
+        thought,
+        finalAnswer: text.trim()
+      }
+    }
+
+    const toolCalls: ToolCall[] = []
+
+    if (text.includes('```')) {
+      const actionBlocks = text
+        .trim()
+        .split(/```(?:json)?/)
+        .filter((block: string) => block.includes(ACTION_TAG))
+
+      actionBlocks.forEach((block: string) => {
+        try {
+          const { action, action_input } = JSON.parse(block.trim())
+
+          if (!action || typeof action !== 'string') {
+            console.error('Invalid tool call: missing or invalid action field')
+
+            return
+          }
+
+          toolCalls.push({
+            id: `call_${Math.random().toString(36).slice(2)}`,
+            type: 'function',
+            function: {
+              name: action,
+              arguments: typeof action_input === 'string' ? action_input : JSON.stringify(action_input || {})
+            }
+          })
+        } catch (error) {
+          console.error('Failed to parse tool call JSON:', error)
+        }
+      })
+    }
+
+    return {
+      toolCalls: toolCalls,
+      thought,
+      finalAnswer: text.trim()
+    }
+  } catch (error) {
+    console.error('Failed to organize tool calls:', error)
+
+    return {
+      toolCalls: [],
+      thought: '',
+      finalAnswer: text
+    }
+  }
+}
+
+export const runReActLoop = async ({
+  step,
+  tools,
+  vm,
+  chatMethod,
+  options,
+  system,
+  llm
+}: {
+  step: any
+  tools: Tool[]
+  vm: any
+  chatMethod: any
+  options: any
+  system: string
+  llm: any
+}) => {
+  const toolCallsResult = []
+  const { toolCalls, thought, finalAnswer } = await organizeToolCalls(step.content[0].text)
+
+  for (const toolCall of toolCalls) {
+    const tool = tools[toolCall.function.name]
+    if (tool) {
+      const result = await tool.execute(JSON.parse(toolCall.function.arguments))
+      toolCallsResult.push(result)
+    }
+  }
+
+  if (toolCallsResult.length > 0) {
+    const lastmessage = step.content[0]
+    lastmessage.text =
+      lastmessage.text +
+      `\n Observation: ${toolCallsResult.map((item: any) => item.content.map((item: any) => item.text).join('\n')).join('\n')}`
+
+    const result =  chatMethod({
+      system: system,
+      model: llm,
+      tools: tools as ToolSet,
+      stopWhen: stepCountIs(options.maxSteps),
+      onStepFinish: async (stepCopy) => {
+        await runReActLoop({ step: stepCopy, tools, vm, chatMethod, options, system: system, llm: llm })
+      },
+      prompt: lastmessage.text,
+    })
+
+    for await (const part of result.fullStream) {
+      part.text &&  options.handler.onData({
+        type: 'markdown',
+        delta: part.text,
+      })
+    }
+
+  } else {
+    return finalAnswer
+  }
+}