ipcHandlers.ts

// LangGraph streaming handler for IPC transport
  // Handles streaming from LangGraph assistant over dedicated event channels
  ipcMain.handle('lg/stream/start', async (event, payload) => {
    const {
      runId,
      eventChannel,
      input,
      context,
      command,
      config,
    }: {
      runId: string;
      eventChannel: string;
      input: any;
      context?: { model?: string; webSearch?: boolean; threadId?: string };
      command?: any;
      config?: any;
    } = payload;

    log.info(`Starting LangGraph stream ${runId} on channel ${eventChannel}`);
    log.debug(`Context:`, context);

    try {
      // Initialize LangGraph client with production credentials
      const client = new Client({
        apiUrl: 'http://localhost:2024/',
      });

      // Get the first assistant
      const assistants = await client.assistants.search({
        metadata: null,
        offset: 0,
        limit: 5,
      });

      if (!assistants || assistants.length === 0) {
        throw new Error('No assistants found');
      }

      const assistant = assistants[0];
      log.info(`Using assistant: ${assistant.assistant_id}`);

      // Use existing thread or create new one
      const threadId = context?.threadId;
      log.info(`Thread ID: ${threadId || 'creating new thread'}`);

      // Build config with context as configurable parameters
      const streamConfig = {
        ...config,
        configurable: {
          ...config?.configurable,
          ...(context && {
            model: context.model,
            webSearch: context.webSearch,
          }),
        },
      };

      // Start streaming from LangGraph - handle null vs string threadId properly
      const streamResponse = threadId
        ? client.runs.stream(threadId, assistant.assistant_id, {
            input: input || {},
            config: streamConfig,
            streamMode: 'messages-tuple',
          })
        : client.runs.stream(null, assistant.assistant_id, {
            input: input || {},
            config: streamConfig,
            streamMode: 'messages',
          });

      let newThreadId: string | null = null;

      // Stream events back to the renderer via the dedicated event channel
      // eslint-disable-next-line no-restricted-syntax
      for await (const chunk of streamResponse) {
        log.debug(`Stream chunk:`, chunk);

        // Capture thread_id from metadata events
        if (chunk.event === 'metadata' && chunk.data?.thread_id) {
          newThreadId = chunk.data.thread_id;
          log.info(`Thread ID from stream: ${newThreadId}`);
        }

        // Forward the stream event to the renderer
        event.sender.send(eventChannel, {
          event: chunk.event,
          data: chunk.data,
          id: runId,
        });
      }

      // Send thread_id back to renderer so it can be used for future messages
      if (newThreadId) {
        event.sender.send(eventChannel, {
          event: 'thread',
          data: { thread_id: newThreadId },
          id: runId,
        });
      }

      log.info(`Stream ${runId} completed successfully`);
      return { success: true, threadId: newThreadId };
    } catch (error) {
      log.error(`Error in LangGraph stream ${runId}:`, error);

      // Send error event to renderer
      event.sender.send(eventChannel, {
        event: 'error',
        data: {
          error: error instanceof Error ? error.name : 'StreamError',
          message: error instanceof Error ? error.message : 'Unknown error',
        },
        id: runId,
      });

      throw error;
    }
  });

  // LangGraph stream cancellation handler
  ipcMain.handle('lg/stream/cancel', async (event, payload) => {
    const { runId }: { runId: string } = payload;
    log.info(`Cancellation requested for stream ${runId}`);

    // TODO: Implement cancellation logic if LangGraph SDK supports it
    // For now, just acknowledge the cancellation request
    return { success: true, runId };
  });