refactor: data flow inversion, IPC contract fixes, and frontend adaptation

Data flow inversion: - Add getHistoryForAgent() in conversations.ts — Agent now reads history directly from SQLite instead of receiving it from frontend - Remove historyMessages parameter from agent:runStream IPC chain, add maxHistoryRounds for configurable context window - Frontend (useAIChat.ts) pre-creates conversation before Agent call, removes history message collection logic IPC fixes: - Restore llm:chat and llm:chatStream handlers using pi-ai's completeSimple/streamSimple (P0-1: SQLLab compatibility) - Send agent:complete with aborted flag in all abort paths to prevent hanging Promises in renderer (P0-2) - Fix createConversation parameter order to match preload/data layer contract: (sessionId, title?) instead of (title, sessionId?) (P1) Preload/frontend: - Update preload API signatures and type declarations - Use shared types for TokenUsage and AgentRuntimeStatus
2026-05-03 03:31:18 +08:00 · 2026-02-26 21:06:27 +08:00
parent 1b059a52cc
commit 07fffb0112
6 changed files with 219 additions and 209 deletions
--- a/electron/main/ai/conversations.ts
+++ b/electron/main/ai/conversations.ts
@@ -363,3 +363,27 @@ export function deleteMessage(messageId: string): boolean {
  const result = db.prepare('DELETE FROM ai_message WHERE id = ?').run(messageId)
  return result.changes > 0
 }
+
+// ==================== Agent 专用 ====================
+
+/**
+ * 为 Agent 提供对话历史
+ *
+ * 返回简化的 {role, content} 格式，按时间升序排列。
+ * @param conversationId 对话 ID
+ * @param maxMessages 最大返回条数（取最近 N 条）
+ */
+export function getHistoryForAgent(
+  conversationId: string,
+  maxMessages?: number
+): Array<{ role: 'user' | 'assistant'; content: string }> {
+  const messages = getMessages(conversationId)
+  const filtered = messages
+    .filter((m) => (m.role === 'user' || m.role === 'assistant') && m.content?.trim())
+    .map((m) => ({ role: m.role, content: m.content }))
+
+  if (maxMessages && filtered.length > maxMessages) {
+    return filtered.slice(-maxMessages)
+  }
+  return filtered
+}
--- a/electron/main/ipc/ai.ts
+++ b/electron/main/ipc/ai.ts
@@ -8,6 +8,8 @@ import * as rag from '../ai/rag'
 import { aiLogger } from '../ai/logger'
 import { getLogsDir } from '../paths'
 import { Agent, type AgentStreamChunk, type PromptConfig } from '../ai/agent'
+import { getActiveConfig, buildPiModel } from '../ai/llm'
+import { completeSimple, streamSimple, type TextContent as PiTextContent } from '@mariozechner/pi-ai'
 import { t } from '../i18n'
 import type { ToolContext } from '../ai/tools/types'
 import type { IpcContext } from './types'
@@ -115,23 +117,16 @@ export function registerAIHandlers({ win }: IpcContext): void {

  /**
   * 创建新的 AI 对话
+   * 参数契约与 preload / 数据层保持一致：(sessionId, title?)
   */
-  ipcMain.handle(
-    'ai:createConversation',
-    async (
-      _,
-      title: string,
-      sessionId?: string,
-      dataSource?: { type: 'chat' | 'member'; id: string; name?: string }
-    ) => {
-      try {
-        return aiConversations.createConversation(title, sessionId, dataSource)
-      } catch (error) {
-        console.error('Failed to create AI conversation:', error)
-        throw error
-      }
+  ipcMain.handle('ai:createConversation', async (_, sessionId: string, title?: string) => {
+    try {
+      return aiConversations.createConversation(sessionId, title)
+    } catch (error) {
+      console.error('Failed to create AI conversation:', error)
+      throw error
    }
-  )
+  })

  /**
   * 获取所有 AI 对话列表
@@ -405,13 +400,11 @@ export function registerAIHandlers({ win }: IpcContext): void {
    async (_, provider: llm.LLMProvider, apiKey: string, baseUrl?: string, model?: string) => {
      console.log('[LLM:validateApiKey] Validating:', { provider, baseUrl, model, apiKeyLength: apiKey?.length })
      try {
-        const service = llm.createLLMService({ provider, apiKey, baseUrl, model })
-        const result = await service.validateApiKey()
+        const result = await llm.validateApiKey(provider, apiKey, baseUrl, model)
        console.log('[LLM:validateApiKey] Result:', result)
-        return { success: result.success, error: result.error }
+        return result
      } catch (error) {
        console.error('[LLM:validateApiKey] Validation failed:', error)
-        // 提取有意义的错误信息
        const errorMessage = error instanceof Error ? error.message : String(error)
        return { success: false, error: errorMessage }
      }
@@ -425,75 +418,124 @@ export function registerAIHandlers({ win }: IpcContext): void {
    return llm.hasActiveConfig()
  })

-  /**
-   * 发送 LLM 聊天请求（非流式）
-   */
-  ipcMain.handle('llm:chat', async (_, messages: llm.ChatMessage[], options?: llm.ChatOptions) => {
-    aiLogger.info('IPC', 'Non-streaming LLM request received', {
-      messagesCount: messages.length,
-      firstMsgRole: messages[0]?.role,
-      firstMsgContentLen: messages[0]?.content?.length,
-      options,
-    })
-    try {
-      const response = await llm.chat(messages, options)
-      aiLogger.info('IPC', 'Non-streaming LLM request succeeded', { responseLength: response.length })
-      return { success: true, content: response }
-    } catch (error) {
-      aiLogger.error('IPC', 'Non-streaming LLM request failed', { error: String(error) })
-      console.error('LLM chat failed:', error)
-      return { success: false, error: String(error) }
-    }
-  })
+  // ==================== LLM 直接调用 API（SQLLab 等非 Agent 场景使用） ====================

  /**
-   * 发送 LLM 聊天请求（流式）
-   * 使用 IPC 事件发送流式数据
+   * 非流式 LLM 调用
+   */
+  ipcMain.handle(
+    'llm:chat',
+    async (
+      _,
+      messages: Array<{ role: string; content: string }>,
+      options?: { temperature?: number; maxTokens?: number }
+    ) => {
+      try {
+        const activeConfig = getActiveConfig()
+        if (!activeConfig) {
+          return { success: false, error: t('llm.notConfigured') }
+        }
+        const piModel = buildPiModel(activeConfig)
+        const now = Date.now()
+        const systemMsg = messages.find((m) => m.role === 'system')
+        const nonSystemMsgs = messages.filter((m) => m.role !== 'system')
+
+        const result = await completeSimple(
+          piModel,
+          {
+            systemPrompt: systemMsg?.content,
+            messages: nonSystemMsgs.map((m) => ({
+              role: m.role as 'user' | 'assistant',
+              content: m.content,
+              timestamp: now,
+            })),
+          },
+          {
+            apiKey: activeConfig.apiKey,
+            temperature: options?.temperature,
+            maxTokens: options?.maxTokens,
+          }
+        )
+
+        const content = result.content
+          .filter((item): item is PiTextContent => item.type === 'text')
+          .map((item) => item.text)
+          .join('')
+
+        return { success: true, content }
+      } catch (error) {
+        aiLogger.error('IPC', 'llm:chat error', { error: String(error) })
+        return { success: false, error: String(error) }
+      }
+    }
+  )
+
+  /**
+   * 流式 LLM 调用（SQLLab AI 生成 / 结果总结等场景使用）
   */
  ipcMain.handle(
    'llm:chatStream',
-    async (_, requestId: string, messages: llm.ChatMessage[], options?: llm.ChatOptions) => {
-      aiLogger.info('IPC', `Streaming chat request received: ${requestId}`, {
-        messagesCount: messages.length,
-        options,
-      })
-
+    async (
+      _,
+      requestId: string,
+      messages: Array<{ role: string; content: string }>,
+      options?: { temperature?: number; maxTokens?: number }
+    ) => {
      try {
-        const generator = llm.chatStream(messages, options)
-        aiLogger.info('IPC', `Stream generator created: ${requestId}`)
+        const activeConfig = getActiveConfig()
+        if (!activeConfig) {
+          return { success: false, error: t('llm.notConfigured') }
+        }
+        const piModel = buildPiModel(activeConfig)
+        const now = Date.now()
+        const systemMsg = messages.find((m) => m.role === 'system')
+        const nonSystemMsgs = messages.filter((m) => m.role !== 'system')

-        // 异步处理流式响应
+        const eventStream = streamSimple(
+          piModel,
+          {
+            systemPrompt: systemMsg?.content,
+            messages: nonSystemMsgs.map((m) => ({
+              role: m.role as 'user' | 'assistant',
+              content: m.content,
+              timestamp: now,
+            })),
+          },
+          {
+            apiKey: activeConfig.apiKey,
+            temperature: options?.temperature,
+            maxTokens: options?.maxTokens,
+          }
+        )
+
+        // 异步消费流，通过事件发送 chunks
        ;(async () => {
-          let chunkIndex = 0
          try {
-            aiLogger.info('IPC', `Iterating stream response: ${requestId}`)
-            for await (const chunk of generator) {
-              chunkIndex++
-              aiLogger.debug('IPC', `Sending chunk #${chunkIndex}: ${requestId}`, {
-                contentLength: chunk.content?.length,
-                isFinished: chunk.isFinished,
-                finishReason: chunk.finishReason,
-              })
-              win.webContents.send('llm:streamChunk', { requestId, chunk })
+            for await (const event of eventStream) {
+              if (event.type === 'text_delta') {
+                win.webContents.send('llm:streamChunk', {
+                  requestId,
+                  chunk: { content: event.delta, isFinished: false },
+                })
+              }
            }
-            aiLogger.info('IPC', `Stream response completed: ${requestId}`, { totalChunks: chunkIndex })
-          } catch (error) {
-            aiLogger.error('IPC', `Stream response error: ${requestId}`, {
-              error: String(error),
-              chunkIndex,
-            })
            win.webContents.send('llm:streamChunk', {
              requestId,
-              chunk: { content: '', isFinished: true, finishReason: 'error' },
+              chunk: { content: '', isFinished: true, finishReason: 'stop' },
+            })
+          } catch (error) {
+            aiLogger.error('IPC', 'llm:chatStream stream error', { requestId, error: String(error) })
+            win.webContents.send('llm:streamChunk', {
+              requestId,
              error: String(error),
+              chunk: { content: '', isFinished: true, finishReason: 'error' },
            })
          }
        })()

        return { success: true }
      } catch (error) {
-        aiLogger.error('IPC', `Failed to create stream request: ${requestId}`, { error: String(error) })
-        console.error('LLM streaming chat failed:', error)
+        aiLogger.error('IPC', 'llm:chatStream error', { error: String(error) })
        return { success: false, error: String(error) }
      }
    }
@@ -504,10 +546,11 @@ export function registerAIHandlers({ win }: IpcContext): void {
  /**
   * 执行 Agent 对话（流式）
   * Agent 会自动调用工具并返回最终结果
-   * @param historyMessages 对话历史（可选，用于上下文关联）
+   * Agent 通过 context.conversationId 从 SQLite 读取对话历史（数据流倒置）
   * @param chatType 聊天类型（'group' | 'private'）
   * @param promptConfig 用户自定义提示词配置（可选）
   * @param locale 语言设置（可选，默认 'zh-CN'）
+   * @param maxHistoryRounds 前端用户配置的最大历史轮数（可选，每轮 = user + assistant = 2 条）
   */
  ipcMain.handle(
    'agent:runStream',
@@ -516,35 +559,36 @@ export function registerAIHandlers({ win }: IpcContext): void {
      requestId: string,
      userMessage: string,
      context: ToolContext,
-      historyMessages?: Array<{ role: 'user' | 'assistant'; content: string }>,
      chatType?: 'group' | 'private',
      promptConfig?: PromptConfig,
-      locale?: string
+      locale?: string,
+      maxHistoryRounds?: number
    ) => {
      aiLogger.info('IPC', `Agent stream request received: ${requestId}`, {
        userMessage: userMessage.slice(0, 100),
        sessionId: context.sessionId,
-        historyLength: historyMessages?.length ?? 0,
+        conversationId: context.conversationId,
        chatType: chatType ?? 'group',
        hasPromptConfig: !!promptConfig,
      })

      try {
-        // 创建 AbortController 并存储
        const abortController = new AbortController()
        activeAgentRequests.set(requestId, abortController)

-        // 转换历史消息格式
-        const formattedHistory =
-          historyMessages?.map((msg) => ({
-            role: msg.role as 'user' | 'assistant',
-            content: msg.content,
-          })) ?? []
+        const activeAIConfig = getActiveConfig()
+        if (!activeAIConfig) {
+          return { success: false, error: t('llm.notConfigured') }
+        }
+        const piModel = buildPiModel(activeAIConfig)
+
+        const contextHistoryLimit = maxHistoryRounds ? maxHistoryRounds * 2 : undefined

        const agent = new Agent(
          context,
-          { abortSignal: abortController.signal },
-          formattedHistory,
+          piModel,
+          activeAIConfig.apiKey,
+          { abortSignal: abortController.signal, contextHistoryLimit },
          chatType ?? 'group',
          promptConfig,
          locale ?? 'zh-CN'
@@ -568,9 +612,18 @@ export function registerAIHandlers({ win }: IpcContext): void {
              win.webContents.send('agent:streamChunk', { requestId, chunk })
            })

-            // 如果已中止，不发送完成信息
            if (abortController.signal.aborted) {
-              aiLogger.info('IPC', `Agent aborted, skipping completion: ${requestId}`)
+              aiLogger.info('IPC', `Agent aborted: ${requestId}`)
+              win.webContents.send('agent:complete', {
+                requestId,
+                result: {
+                  content: result.content,
+                  toolsUsed: result.toolsUsed,
+                  toolRounds: result.toolRounds,
+                  totalUsage: result.totalUsage,
+                  aborted: true,
+                },
+              })
              return
            }

@@ -592,9 +645,12 @@ export function registerAIHandlers({ win }: IpcContext): void {
              totalUsage: result.totalUsage,
            })
          } catch (error) {
-            // 如果是中止错误，不报告为错误
            if (error instanceof Error && error.name === 'AbortError') {
-              aiLogger.info('IPC', `Agent request aborted: ${requestId}`)
+              aiLogger.info('IPC', `Agent request aborted (error): ${requestId}`)
+              win.webContents.send('agent:complete', {
+                requestId,
+                result: { content: '', toolsUsed: [], toolRounds: 0, aborted: true },
+              })
              return
            }
            const friendlyError = formatAIError(error)
--- a/electron/main/ipc/chat.ts
+++ b/electron/main/ipc/chat.ts
@@ -295,9 +295,7 @@ export function registerChatHandlers(ctx: IpcContext): void {
   */
  ipcMain.handle('chat:deleteSession', async (_, sessionId: string) => {
    try {
-      // 先关闭 Worker 中的数据库连接
      await worker.closeDatabase(sessionId)
-      // 然后删除文件（使用核心模块）
      const result = databaseCore.deleteSession(sessionId)
      return result
    } catch (error) {