LibreChat/api/server/services/ToolService.js

const fs = require('fs');
const path = require('path');
const { zodToJsonSchema } = require('zod-to-json-schema');
const { Calculator } = require('langchain/tools/calculator');
const { tool: toolFn, Tool } = require('@langchain/core/tools');
const {
  Tools,
  ContentTypes,
  imageGenTools,
  actionDelimiter,
  ImageVisionTool,
  openapiToFunction,
  validateAndParseOpenAPISpec,
} = require('librechat-data-provider');
const { processFileURL, uploadImageBuffer } = require('~/server/services/Files/process');
const { loadActionSets, createActionTool, domainParser } = require('./ActionService');
const { recordUsage } = require('~/server/services/Threads');
const { loadTools } = require('~/app/clients/tools/util');
const { redactMessage } = require('~/config/parsers');
const { sleep } = require('~/server/utils');
const { logger } = require('~/config');

const filteredTools = new Set([
  'ChatTool.js',
  'CodeSherpa.js',
  'CodeSherpaTools.js',
  'E2BTools.js',
  'extractionChain.js',
]);

/**
 * Loads and formats tools from the specified tool directory.
 *
 * The directory is scanned for JavaScript files, excluding any files in the filter set.
 * For each file, it attempts to load the file as a module and instantiate a class, if it's a subclass of `StructuredTool`.
 * Each tool instance is then formatted to be compatible with the OpenAI Assistant.
 * Additionally, instances of LangChain Tools are included in the result.
 *
 * @param {object} params - The parameters for the function.
 * @param {string} params.directory - The directory path where the tools are located.
 * @param {Array<string>} [params.adminFilter=[]] - Array of admin-defined tool keys to exclude from loading.
 * @param {Array<string>} [params.adminIncluded=[]] - Array of admin-defined tool keys to include from loading.
 * @returns {Record<string, FunctionTool>} An object mapping each tool's plugin key to its instance.
 */
function loadAndFormatTools({ directory, adminFilter = [], adminIncluded = [] }) {
  const filter = new Set([...adminFilter, ...filteredTools]);
  const included = new Set(adminIncluded);
  const tools = [];
  /* Structured Tools Directory */
  const files = fs.readdirSync(directory);

  if (included.size > 0 && adminFilter.length > 0) {
    logger.warn(
      'Both `includedTools` and `filteredTools` are defined; `filteredTools` will be ignored.',
    );
  }

  for (const file of files) {
    const filePath = path.join(directory, file);
    if (!file.endsWith('.js') || (filter.has(file) && included.size === 0)) {
      continue;
    }

    let ToolClass = null;
    try {
      ToolClass = require(filePath);
    } catch (error) {
      logger.error(`[loadAndFormatTools] Error loading tool from ${filePath}:`, error);
      continue;
    }

    if (!ToolClass || !(ToolClass.prototype instanceof Tool)) {
      continue;
    }

    let toolInstance = null;
    try {
      toolInstance = new ToolClass({ override: true });
    } catch (error) {
      logger.error(
        `[loadAndFormatTools] Error initializing \`${file}\` tool; if it requires authentication, is the \`override\` field configured?`,
        error,
      );
      continue;
    }

    if (!toolInstance) {
      continue;
    }

    if (filter.has(toolInstance.name) && included.size === 0) {
      continue;
    }

    if (included.size > 0 && !included.has(file) && !included.has(toolInstance.name)) {
      continue;
    }

    const formattedTool = formatToOpenAIAssistantTool(toolInstance);
    tools.push(formattedTool);
  }

  /** Basic Tools; schema: { input: string } */
  const basicToolInstances = [new Calculator()];
  for (const toolInstance of basicToolInstances) {
    const formattedTool = formatToOpenAIAssistantTool(toolInstance);
    tools.push(formattedTool);
  }

  tools.push(ImageVisionTool);

  return tools.reduce((map, tool) => {
    map[tool.function.name] = tool;
    return map;
  }, {});
}

/**
 * Formats a `StructuredTool` instance into a format that is compatible
 * with OpenAI's ChatCompletionFunctions. It uses the `zodToJsonSchema`
 * function to convert the schema of the `StructuredTool` into a JSON
 * schema, which is then used as the parameters for the OpenAI function.
 *
 * @param {StructuredTool} tool - The StructuredTool to format.
 * @returns {FunctionTool} The OpenAI Assistant Tool.
 */
function formatToOpenAIAssistantTool(tool) {
  return {
    type: Tools.function,
    [Tools.function]: {
      name: tool.name,
      description: tool.description,
      parameters: zodToJsonSchema(tool.schema),
    },
  };
}

/**
 * Processes the required actions by calling the appropriate tools and returning the outputs.
 * @param {OpenAIClient} client - OpenAI or StreamRunManager Client.
 * @param {RequiredAction} requiredActions - The current required action.
 * @returns {Promise<ToolOutput>} The outputs of the tools.
 */
const processVisionRequest = async (client, currentAction) => {
  if (!client.visionPromise) {
    return {
      tool_call_id: currentAction.toolCallId,
      output: 'No image details found.',
    };
  }

  /** @type {ChatCompletion | undefined} */
  const completion = await client.visionPromise;
  if (completion.usage) {
    recordUsage({
      user: client.req.user.id,
      model: client.req.body.model,
      conversationId: (client.responseMessage ?? client.finalMessage).conversationId,
      ...completion.usage,
    });
  }
  const output = completion?.choices?.[0]?.message?.content ?? 'No image details found.';
  return {
    tool_call_id: currentAction.toolCallId,
    output,
  };
};

/**
 * Processes return required actions from run.
 * @param {OpenAIClient | StreamRunManager} client - OpenAI (legacy) or StreamRunManager Client.
 * @param {RequiredAction[]} requiredActions - The required actions to submit outputs for.
 * @returns {Promise<ToolOutputs>} The outputs of the tools.
 */
async function processRequiredActions(client, requiredActions) {
  logger.debug(
    `[required actions] user: ${client.req.user.id} | thread_id: ${requiredActions[0].thread_id} | run_id: ${requiredActions[0].run_id}`,
    requiredActions,
  );
  const tools = requiredActions.map((action) => action.tool);
  const loadedTools = await loadTools({
    user: client.req.user.id,
    model: client.req.body.model ?? 'gpt-4o-mini',
    tools,
    functions: true,
    options: {
      processFileURL,
      req: client.req,
      uploadImageBuffer,
      openAIApiKey: client.apiKey,
      fileStrategy: client.req.app.locals.fileStrategy,
      returnMetadata: true,
    },
    skipSpecs: true,
  });

  const ToolMap = loadedTools.reduce((map, tool) => {
    map[tool.name] = tool;
    return map;
  }, {});

  const promises = [];

  /** @type {Action[]} */
  let actionSets = [];
  let isActionTool = false;
  const ActionToolMap = {};
  const ActionBuildersMap = {};

  for (let i = 0; i < requiredActions.length; i++) {
    const currentAction = requiredActions[i];
    if (currentAction.tool === ImageVisionTool.function.name) {
      promises.push(processVisionRequest(client, currentAction));
      continue;
    }
    let tool = ToolMap[currentAction.tool] ?? ActionToolMap[currentAction.tool];

    const handleToolOutput = async (output) => {
      requiredActions[i].output = output;

      /** @type {FunctionToolCall & PartMetadata} */
      const toolCall = {
        function: {
          name: currentAction.tool,
          arguments: JSON.stringify(currentAction.toolInput),
          output,
        },
        id: currentAction.toolCallId,
        type: 'function',
        progress: 1,
        action: isActionTool,
      };

      const toolCallIndex = client.mappedOrder.get(toolCall.id);

      if (imageGenTools.has(currentAction.tool)) {
        const imageOutput = output;
        toolCall.function.output = `${currentAction.tool} displayed an image. All generated images are already plainly visible, so don't repeat the descriptions in detail. Do not list download links as they are available in the UI already. The user may download the images by clicking on them, but do not mention anything about downloading to the user.`;

        // Streams the "Finished" state of the tool call in the UI
        client.addContentData({
          [ContentTypes.TOOL_CALL]: toolCall,
          index: toolCallIndex,
          type: ContentTypes.TOOL_CALL,
        });

        await sleep(500);

        /** @type {ImageFile} */
        const imageDetails = {
          ...imageOutput,
          ...currentAction.toolInput,
        };

        const image_file = {
          [ContentTypes.IMAGE_FILE]: imageDetails,
          type: ContentTypes.IMAGE_FILE,
          // Replace the tool call output with Image file
          index: toolCallIndex,
        };

        client.addContentData(image_file);

        // Update the stored tool call
        client.seenToolCalls && client.seenToolCalls.set(toolCall.id, toolCall);

        return {
          tool_call_id: currentAction.toolCallId,
          output: toolCall.function.output,
        };
      }

      client.seenToolCalls && client.seenToolCalls.set(toolCall.id, toolCall);
      client.addContentData({
        [ContentTypes.TOOL_CALL]: toolCall,
        index: toolCallIndex,
        type: ContentTypes.TOOL_CALL,
        // TODO: to append tool properties to stream, pass metadata rest to addContentData
        // result: tool.result,
      });

      return {
        tool_call_id: currentAction.toolCallId,
        output,
      };
    };

    if (!tool) {
      // throw new Error(`Tool ${currentAction.tool} not found.`);

      if (!actionSets.length) {
        actionSets =
          (await loadActionSets({
            assistant_id: client.req.body.assistant_id,
          })) ?? [];
      }

      let actionSet = null;
      let currentDomain = '';
      for (let action of actionSets) {
        const domain = await domainParser(client.req, action.metadata.domain, true);
        if (currentAction.tool.includes(domain)) {
          currentDomain = domain;
          actionSet = action;
          break;
        }
      }

      if (!actionSet) {
        // TODO: try `function` if no action set is found
        // throw new Error(`Tool ${currentAction.tool} not found.`);
        continue;
      }

      let builders = ActionBuildersMap[actionSet.metadata.domain];

      if (!builders) {
        const validationResult = validateAndParseOpenAPISpec(actionSet.metadata.raw_spec);
        if (!validationResult.spec) {
          throw new Error(
            `Invalid spec: user: ${client.req.user.id} | thread_id: ${requiredActions[0].thread_id} | run_id: ${requiredActions[0].run_id}`,
          );
        }
        const { requestBuilders } = openapiToFunction(validationResult.spec);
        ActionToolMap[actionSet.metadata.domain] = requestBuilders;
        builders = requestBuilders;
      }

      const functionName = currentAction.tool.replace(`${actionDelimiter}${currentDomain}`, '');

      const requestBuilder = builders[functionName];

      if (!requestBuilder) {
        // throw new Error(`Tool ${currentAction.tool} not found.`);
        continue;
      }

      tool = await createActionTool({ action: actionSet, requestBuilder });
      isActionTool = !!tool;
      ActionToolMap[currentAction.tool] = tool;
    }

    if (currentAction.tool === 'calculator') {
      currentAction.toolInput = currentAction.toolInput.input;
    }

    const handleToolError = (error) => {
      logger.error(
        `tool_call_id: ${currentAction.toolCallId} | Error processing tool ${currentAction.tool}`,
        error,
      );
      return {
        tool_call_id: currentAction.toolCallId,
        output: `Error processing tool ${currentAction.tool}: ${redactMessage(error.message, 256)}`,
      };
    };

    try {
      const promise = tool
        ._call(currentAction.toolInput)
        .then(handleToolOutput)
        .catch(handleToolError);
      promises.push(promise);
    } catch (error) {
      const toolOutputError = handleToolError(error);
      promises.push(Promise.resolve(toolOutputError));
    }
  }

  return {
    tool_outputs: await Promise.all(promises),
  };
}

/**
 * Processes the runtime tool calls and returns a combined toolMap.
 * @param {Object} params - Run params containing user and request information.
 * @param {ServerRequest} params.req - The request object.
 * @param {string} params.agent_id - The agent ID.
 * @param {Agent['tools']} params.tools - The agent's available tools.
 * @param {Agent['tool_resources']} params.tool_resources - The agent's available tool resources.
 * @param {string | undefined} [params.openAIApiKey] - The OpenAI API key.
 * @returns {Promise<{ tools?: StructuredTool[]; toolMap?: Record<string, StructuredTool>}>} The combined toolMap.
 */
async function loadAgentTools({ req, agent_id, tools, tool_resources, openAIApiKey }) {
  if (!tools || tools.length === 0) {
    return {};
  }
  const loadedTools = await loadTools({
    user: req.user.id,
    // model: req.body.model ?? 'gpt-4o-mini',
    tools,
    functions: true,
    options: {
      req,
      openAIApiKey,
      tool_resources,
      returnMetadata: true,
      processFileURL,
      uploadImageBuffer,
      fileStrategy: req.app.locals.fileStrategy,
    },
    skipSpecs: true,
  });

  const agentTools = [];
  for (let i = 0; i < loadedTools.length; i++) {
    const tool = loadedTools[i];
    if (tool.name && (tool.name === Tools.execute_code || tool.name === Tools.file_search)) {
      agentTools.push(tool);
      continue;
    }

    const toolInstance = toolFn(
      async (...args) => {
        return tool['_call'](...args);
      },
      {
        name: tool.name,
        description: tool.description,
        schema: tool.schema,
      },
    );

    agentTools.push(toolInstance);
  }

  const ToolMap = loadedTools.reduce((map, tool) => {
    map[tool.name] = tool;
    return map;
  }, {});

  let actionSets = [];
  const ActionToolMap = {};

  for (const toolName of tools) {
    if (!ToolMap[toolName]) {
      if (!actionSets.length) {
        actionSets = (await loadActionSets({ agent_id })) ?? [];
      }

      let actionSet = null;
      let currentDomain = '';
      for (let action of actionSets) {
        const domain = await domainParser(req, action.metadata.domain, true);
        if (toolName.includes(domain)) {
          currentDomain = domain;
          actionSet = action;
          break;
        }
      }

      if (actionSet) {
        const validationResult = validateAndParseOpenAPISpec(actionSet.metadata.raw_spec);
        if (validationResult.spec) {
          const { requestBuilders, functionSignatures, zodSchemas } = openapiToFunction(
            validationResult.spec,
            true,
          );
          const functionName = toolName.replace(`${actionDelimiter}${currentDomain}`, '');
          const functionSig = functionSignatures.find((sig) => sig.name === functionName);
          const requestBuilder = requestBuilders[functionName];
          const zodSchema = zodSchemas[functionName];

          if (requestBuilder) {
            const tool = await createActionTool({
              action: actionSet,
              requestBuilder,
              zodSchema,
              name: toolName,
              description: functionSig.description,
            });
            agentTools.push(tool);
            ActionToolMap[toolName] = tool;
          }
        }
      }
    }
  }

  if (tools.length > 0 && agentTools.length === 0) {
    throw new Error('No tools found for the specified tool calls.');
  }

  const toolMap = { ...ToolMap, ...ActionToolMap };
  return {
    tools: agentTools,
    toolMap,
  };
}

module.exports = {
  loadAgentTools,
  loadAndFormatTools,
  processRequiredActions,
  formatToOpenAIAssistantTool,
};