bolt-diy/app/lib/.server/llm/stream-text.ts

import { convertToCoreMessages, streamText as _streamText, type Message } from 'ai';
import { MAX_TOKENS, PROVIDER_COMPLETION_LIMITS, isReasoningModel, type FileMap } from './constants';
import { getSystemPrompt } from '~/lib/common/prompts/prompts';
import { DEFAULT_MODEL, DEFAULT_PROVIDER, MODIFICATIONS_TAG_NAME, PROVIDER_LIST, WORK_DIR } from '~/utils/constants';
import type { IProviderSetting } from '~/types/model';
import { PromptLibrary } from '~/lib/common/prompt-library';
import { allowedHTMLElements } from '~/utils/markdown';
import { LLMManager } from '~/lib/modules/llm/manager';
import { createScopedLogger } from '~/utils/logger';
import { createFilesContext, extractPropertiesFromMessage } from './utils';
import { discussPrompt } from '~/lib/common/prompts/discuss-prompt';
import type { DesignScheme } from '~/types/design-scheme';

export type Messages = Message[];

export interface StreamingOptions extends Omit<Parameters<typeof _streamText>[0], 'model'> {
  supabaseConnection?: {
    isConnected: boolean;
    hasSelectedProject: boolean;
    credentials?: {
      anonKey?: string;
      supabaseUrl?: string;
    };
  };
}

const logger = createScopedLogger('stream-text');

function getCompletionTokenLimit(modelDetails: any): number {
  // 1. If model specifies completion tokens, use that
  if (modelDetails.maxCompletionTokens && modelDetails.maxCompletionTokens > 0) {
    return modelDetails.maxCompletionTokens;
  }

  // 2. Use provider-specific default
  const providerDefault = PROVIDER_COMPLETION_LIMITS[modelDetails.provider];

  if (providerDefault) {
    return providerDefault;
  }

  // 3. Final fallback to MAX_TOKENS, but cap at reasonable limit for safety
  return Math.min(MAX_TOKENS, 16384);
}

function sanitizeText(text: string): string {
  let sanitized = text.replace(/<div class=\\"__boltThought__\\">.*?<\/div>/s, '');
  sanitized = sanitized.replace(/<think>.*?<\/think>/s, '');
  sanitized = sanitized.replace(/<boltAction type="file" filePath="package-lock\.json">[\s\S]*?<\/boltAction>/g, '');

  return sanitized.trim();
}

export async function streamText(props: {
  messages: Omit<Message, 'id'>[];
  env?: Env;
  options?: StreamingOptions;
  apiKeys?: Record<string, string>;
  files?: FileMap;
  providerSettings?: Record<string, IProviderSetting>;
  promptId?: string;
  contextOptimization?: boolean;
  contextFiles?: FileMap;
  summary?: string;
  messageSliceId?: number;
  chatMode?: 'discuss' | 'build';
  designScheme?: DesignScheme;
}) {
  const {
    messages,
    env: serverEnv,
    options,
    apiKeys,
    files,
    providerSettings,
    promptId,
    contextOptimization,
    contextFiles,
    summary,
    chatMode,
    designScheme,
  } = props;
  let currentModel = DEFAULT_MODEL;
  let currentProvider = DEFAULT_PROVIDER.name;
  let processedMessages = messages.map((message) => {
    const newMessage = { ...message };

    if (message.role === 'user') {
      const { model, provider, content } = extractPropertiesFromMessage(message);
      currentModel = model;
      currentProvider = provider;
      newMessage.content = sanitizeText(content);
    } else if (message.role == 'assistant') {
      newMessage.content = sanitizeText(message.content);
    }

    // Sanitize all text parts in parts array, if present
    if (Array.isArray(message.parts)) {
      newMessage.parts = message.parts.map((part) =>
        part.type === 'text' ? { ...part, text: sanitizeText(part.text) } : part,
      );
    }

    return newMessage;
  });

  const provider = PROVIDER_LIST.find((p) => p.name === currentProvider) || DEFAULT_PROVIDER;
  const staticModels = LLMManager.getInstance().getStaticModelListFromProvider(provider);
  let modelDetails = staticModels.find((m) => m.name === currentModel);

  if (!modelDetails) {
    const modelsList = [
      ...(provider.staticModels || []),
      ...(await LLMManager.getInstance().getModelListFromProvider(provider, {
        apiKeys,
        providerSettings,
        serverEnv: serverEnv as any,
      })),
    ];

    if (!modelsList.length) {
      throw new Error(`No models found for provider ${provider.name}`);
    }

    modelDetails = modelsList.find((m) => m.name === currentModel);

    if (!modelDetails) {
      // Check if it's a Google provider and the model name looks like it might be incorrect
      if (provider.name === 'Google' && currentModel.includes('2.5')) {
        throw new Error(
          `Model "${currentModel}" not found. Gemini 2.5 Pro doesn't exist. Available Gemini models include: gemini-1.5-pro, gemini-2.0-flash, gemini-1.5-flash. Please select a valid model.`,
        );
      }

      // Fallback to first model with warning
      logger.warn(
        `MODEL [${currentModel}] not found in provider [${provider.name}]. Falling back to first model. ${modelsList[0].name}`,
      );
      modelDetails = modelsList[0];
    }
  }

  const dynamicMaxTokens = modelDetails ? getCompletionTokenLimit(modelDetails) : Math.min(MAX_TOKENS, 16384);

  // Additional safety cap - should not be needed with proper completion limits, but kept for safety
  const safeMaxTokens = Math.min(dynamicMaxTokens, 128000);

  logger.info(
    `Max tokens for model ${modelDetails.name} is ${safeMaxTokens} (capped from ${dynamicMaxTokens}) based on model limits`,
  );

  let systemPrompt =
    PromptLibrary.getPropmtFromLibrary(promptId || 'default', {
      cwd: WORK_DIR,
      allowedHtmlElements: allowedHTMLElements,
      modificationTagName: MODIFICATIONS_TAG_NAME,
      designScheme,
      supabase: {
        isConnected: options?.supabaseConnection?.isConnected || false,
        hasSelectedProject: options?.supabaseConnection?.hasSelectedProject || false,
        credentials: options?.supabaseConnection?.credentials || undefined,
      },
    }) ?? getSystemPrompt();

  if (chatMode === 'build' && contextFiles && contextOptimization) {
    const codeContext = createFilesContext(contextFiles, true);

    systemPrompt = `${systemPrompt}

    Below is the artifact containing the context loaded into context buffer for you to have knowledge of and might need changes to fullfill current user request.
    CONTEXT BUFFER:
    ---
    ${codeContext}
    ---
    `;

    if (summary) {
      systemPrompt = `${systemPrompt}
      below is the chat history till now
      CHAT SUMMARY:
      ---
      ${props.summary}
      ---
      `;

      if (props.messageSliceId) {
        processedMessages = processedMessages.slice(props.messageSliceId);
      } else {
        const lastMessage = processedMessages.pop();

        if (lastMessage) {
          processedMessages = [lastMessage];
        }
      }
    }
  }

  const effectiveLockedFilePaths = new Set<string>();

  if (files) {
    for (const [filePath, fileDetails] of Object.entries(files)) {
      if (fileDetails?.isLocked) {
        effectiveLockedFilePaths.add(filePath);
      }
    }
  }

  if (effectiveLockedFilePaths.size > 0) {
    const lockedFilesListString = Array.from(effectiveLockedFilePaths)
      .map((filePath) => `- ${filePath}`)
      .join('\n');
    systemPrompt = `${systemPrompt}

    IMPORTANT: The following files are locked and MUST NOT be modified in any way. Do not suggest or make any changes to these files. You can proceed with the request but DO NOT make any changes to these files specifically:
    ${lockedFilesListString}
    ---
    `;
  } else {
    console.log('No locked files found from any source for prompt.');
  }

  logger.info(`Sending llm call to ${provider.name} with model ${modelDetails.name}`);

  // DEBUG: Log reasoning model detection
  const isReasoning = isReasoningModel(modelDetails.name);
  logger.info(`DEBUG STREAM: Model "${modelDetails.name}" detected as reasoning model: ${isReasoning}`);

  // console.log(systemPrompt, processedMessages);

  // Use maxCompletionTokens for reasoning models (o1, GPT-5), maxTokens for traditional models
  const tokenParams = isReasoning ? { maxCompletionTokens: safeMaxTokens } : { maxTokens: safeMaxTokens };

  // Filter out unsupported parameters for reasoning models
  const filteredOptions =
    isReasoning && options
      ? Object.fromEntries(
          Object.entries(options).filter(
            ([key]) =>
              ![
                'temperature',
                'topP',
                'presencePenalty',
                'frequencyPenalty',
                'logprobs',
                'topLogprobs',
                'logitBias',
              ].includes(key),
          ),
        )
      : options || {};

  // DEBUG: Log filtered options
  logger.info(
    `DEBUG STREAM: Options filtering for model "${modelDetails.name}":`,
    JSON.stringify(
      {
        isReasoning,
        originalOptions: options || {},
        filteredOptions,
        originalOptionsKeys: options ? Object.keys(options) : [],
        filteredOptionsKeys: Object.keys(filteredOptions),
        removedParams: options ? Object.keys(options).filter((key) => !(key in filteredOptions)) : [],
      },
      null,
      2,
    ),
  );

  const streamParams = {
    model: provider.getModelInstance({
      model: modelDetails.name,
      serverEnv,
      apiKeys,
      providerSettings,
    }),
    system: chatMode === 'build' ? systemPrompt : discussPrompt(),
    ...tokenParams,
    messages: convertToCoreMessages(processedMessages as any),
    ...filteredOptions,

    // Set temperature to 1 for reasoning models (required by OpenAI API)
    ...(isReasoning ? { temperature: 1 } : {}),
  };

  // DEBUG: Log final streaming parameters
  logger.info(
    `DEBUG STREAM: Final streaming params for model "${modelDetails.name}":`,
    JSON.stringify(
      {
        hasTemperature: 'temperature' in streamParams,
        hasMaxTokens: 'maxTokens' in streamParams,
        hasMaxCompletionTokens: 'maxCompletionTokens' in streamParams,
        paramKeys: Object.keys(streamParams).filter((key) => !['model', 'messages', 'system'].includes(key)),
        streamParams: Object.fromEntries(
          Object.entries(streamParams).filter(([key]) => !['model', 'messages', 'system'].includes(key)),
        ),
      },
      null,
      2,
    ),
  );

  return await _streamText(streamParams);
}