LibreChat/api/server/controllers/tools.js
Danny Avila 0dbbf7de04
🔎 feat: Native Web Search with Citation References (#7516)
* WIP: search tool integration

* WIP: Add web search capabilities and API key management to agent actions

* WIP: web search capability to agent configuration and selection

* WIP: Add web search capability to backend agent configuration

* WIP: add web search option to default agent form values

* WIP: add attachments for web search

* feat: add plugin for processing web search citations

* WIP: first pass, Citation UI

* chore: remove console.log

* feat: Add AnimatedTabs component for tabbed UI functionality

* refactor: AnimatedTabs component with CSS animations and stable ID generation

* WIP example content

* feat: SearchContext for managing search results apart from MessageContext

* feat: Enhance AnimatedTabs with underline animation and state management

* WIP: first pass, Implement dynamic tab functionality in Sources component with search results integration

* fix: Update class names for improved styling in Sources and AnimatedTabs components

* feat: Improve styling and layout in Sources component with enhanced button and item designs

* feat: Refactor Sources component to integrate OGDialog for source display and improve layout

* style: Update background color in SourceItem and SourcesGroup components for improved visibility

* refactor: Sources component to enhance SourceItem structure and improve favicon handling

* style: Adjust font size of domain text in SourceItem for better readability

* feat: Add localization for citation source and details in CompositeCitation component

* style: add theming to Citation components

* feat: Enhance SourceItem component with dialog support and improved hovercard functionality

* feat: Add localization for sources tab and image alt text in Sources component

* style: Replace divs with spans for better semantic structure in CompositeCitation and Citation components

* refactor: Sources component to use useMemo for tab generation and improve performance

* chore: bump @librechat/agents to v2.4.318

* chore: update search result types

* fix: search results retrieval in ContentParts component, re-render attachments when expected

* feat: update sources style/types to use latest search result structure

* style: enhance Dialog (expanded) SourceItem component with link wrapping and improved styling

* style: update ImageItem component styling for improved title visibility

* refactor: remove SourceItemBase component and adjust SourceItem layout for improved styling

* chore: linting twcss order

* fix: prevent FileAttachment from rendering search attachments

* fix: append underscore to responseMessageId for unique identification to prevent mapping of previous latest message's attachments

* chore: remove unused parameter 'useSpecs' from loadTools function

* chore: twcss order

* WIP: WebSearch Tool UI

* refactor: add limit parameter to StackedFavicons for customizable source display

* refactor: optimize search results memoization by making more granular and separate conerns

* refactor: integrated StackedFavicons to WebSearch mid-run

* chore: bump @librechat/agents to expose handleToolCallChunks

* chore: use typedefs from dedicated file instead of defining them in AgentClient module

* WIP: first pass, search progress results

* refactor: move createOnSearchResults function to a dedicated search module

* chore: bump @librechat/agents to v2.4.320

* WIP: first pass, search results processed UX

* refactor: consolidate context variables in createOnSearchResults function

* chore: bump @librechat/agents to v2.4.321

* feat: add guidelines for web search tool response formatting in loadTools function

* feat: add isLast prop to Part component and update WebSearch logic for improved state handling

* style: update Hovercard styles for improved UI consistency

* feat: export FaviconImage component for improved accessibility in other modules

* refactor: export getCleanDomain function and use FaviconImage in Citation component for improved source representation

* refactor: implement SourceHovercard component for consistency and DRY compliance

* fix: replace <p> with <span> for snippet and title in SourceItem and SourceHovercard for consistency

* style: `not-prose`

* style: remove 'not-prose' class for consistency in SourceItem, Citation, and SourceHovercard components, adjust style classes

* refactor: `imageUrl` on hover and prevent duplicate sources

* refactor: enhance SourcesGroup dialog layout and improve source item presentation

* refactor: reorganize Web Components, save in same directory

* feat: add 'news' refType to refTypeMap for citation sources

* style: adjust Hovercard width for improved layout

* refactor: update tool usage guidelines for improved clarity and execution

* chore: linting

* feat: add Web Search badge with initial permissions and local storage logic

* feat: add webSearch support to interface and permissions schemas

* feat: implement Web Search API key management and localization updates

* feat: refactor Web Search API key handling and integrate new search API key form

* fix: remove unnecessary visibility state from FileAttachment component

* feat: update WebSearch component to use Globe icon and localized search label

* feat: enhance ApiKeyDialog with dropdown for reranker selection and update translations

* feat: implement dropdown menus for engine, scraper, and reranker selection in ApiKeyDialog

* chore: linting and add unknown instead of `any` type

* feat: refactor ApiKeyDialog and useAuthSearchTool for improved API key management

* refactor: update ocrSchema to use template literals for default apiKey and baseURL

* feat: add web search configuration and utility functions for environment variable extraction

* fix: ensure filepath is defined before checking its prefix in useAttachmentHandler

* feat: enhance web search functionality with improved configuration and environment variable extraction for authFields

* fix: update auth type in TPluginAction and TUpdateUserPlugins to use Partial<Record<string, string>>

* feat: implement web search authentication verification and enhance webSearchAuth structure

* feat: enhance ephemeral agent handling with new web search capability and type definition

* feat: enhance isEphemeralAgent function to include web search selection

* feat: refactor verifyWebSearchAuth to improve key handling and authentication checks

* feat: implement loadWebSearchAuth function for improved web search authentication handling

* feat: enhance web search authentication with new configuration options and refactor related types

* refactor: rename search engine to search provider and update related localization keys

* feat: update verifyWebSearchAuth to handle multiple authentication types and improve error handling

* feat: update ApiKeyDialog to accept authTypes prop and remove isUserProvided check

* feat: add tests for extractWebSearchEnvVars and loadWebSearchAuth functions

* feat: enhance loadWebSearchAuth to support specific service checks for providers, scrapers, and rerankers

* fix: update web search configuration key and adjust auth result handling in loadTools function

* feat: add new progress key for repeated web searching and update localization

* chore: bump @librechat/agents to 2.4.322

* feat: enhance loadTools function to include ISO time and improve search tool logging

* feat: update StackedFavicons to handle negative start index and improve citation attribution styling and text

* chore: update .gitignore to categorize AI-related files

* fix: mobile responsiveness of sources/citations hovercards

* feat: enhance source display with improved line clamping for better readability

* chore: bump @librechat/agents to v2.4.33

* feat: add handling for image sources in references mapping

* chore: bump librechat-data-provider version to 0.7.84

* chore: bump @librechat/agents version to 2.4.34

* fix: update auth handling to support multiple auth types in tools and allow key configuration in agent panel

* chore: remove redundant agent attribution text from search form

* fix: web search auth uninstall

* refactor: convert CheckboxButton to a forwardRef component and update setValue callback signature

* feat: add triggerRef prop to ApiKeyDialog components for improved dialog control

* feat: integrate triggerRef in CodeInterpreter and WebSearch components for enhanced dialog management

* feat: enhance ApiKeyDialog with additional links for Firecrawl and Jina API key guidance

* feat: implement web search configuration handling in ApiKeyDialog and add tests for dropdown visibility

* fix: update webSearchConfig reference in config route for correct payload assignment

* feat: update ApiKeyDialog to conditionally render sections based on authTypes and modify loadWebSearchAuth to correctly categorize authentication types

* feat: refactor ApiKeyDialog and related tests to use SearchCategories and RerankerTypes enums and remove nested ternaries

* refactor: move ThinkingButton rendering to improve layout consistency in ContentParts

* feat: integrate search context into Markdown component to conditionally include unicodeCitation plugin

* chore: bump @librechat/agents to v2.4.35

* chore: remove unused 18n key

* ci: add WEB_SEARCH permission testing and update AppService tests for new webSearch configuration

* ci: add more comprehensive tests for loadWebSearchAuth to validate authentication handling and authTypes structure

* chore: remove debugging console log from web.spec.ts to clean up test output
2025-05-23 17:40:40 -04:00

242 lines
7.4 KiB
JavaScript

const { nanoid } = require('nanoid');
const { EnvVar } = require('@librechat/agents');
const {
Tools,
AuthType,
Permissions,
ToolCallTypes,
PermissionTypes,
loadWebSearchAuth,
} = require('librechat-data-provider');
const { processFileURL, uploadImageBuffer } = require('~/server/services/Files/process');
const { processCodeOutput } = require('~/server/services/Files/Code/process');
const { createToolCall, getToolCallsByConvo } = require('~/models/ToolCall');
const { loadAuthValues } = require('~/server/services/Tools/credentials');
const { loadTools } = require('~/app/clients/tools/util');
const { checkAccess } = require('~/server/middleware');
const { getMessage } = require('~/models/Message');
const { logger } = require('~/config');
const fieldsMap = {
[Tools.execute_code]: [EnvVar.CODE_API_KEY],
};
const toolAccessPermType = {
[Tools.execute_code]: PermissionTypes.RUN_CODE,
};
/**
* Verifies web search authentication, ensuring each category has at least
* one fully authenticated service.
*
* @param {ServerRequest} req - The request object
* @param {ServerResponse} res - The response object
* @returns {Promise<void>} A promise that resolves when the function has completed
*/
const verifyWebSearchAuth = async (req, res) => {
try {
const userId = req.user.id;
/** @type {TCustomConfig['webSearch']} */
const webSearchConfig = req.app.locals?.webSearch || {};
const result = await loadWebSearchAuth({
userId,
loadAuthValues,
webSearchConfig,
throwError: false,
});
return res.status(200).json({
authenticated: result.authenticated,
authTypes: result.authTypes,
});
} catch (error) {
console.error('Error in verifyWebSearchAuth:', error);
return res.status(500).json({ message: error.message });
}
};
/**
* @param {ServerRequest} req - The request object, containing information about the HTTP request.
* @param {ServerResponse} res - The response object, used to send back the desired HTTP response.
* @returns {Promise<void>} A promise that resolves when the function has completed.
*/
const verifyToolAuth = async (req, res) => {
try {
const { toolId } = req.params;
if (toolId === Tools.web_search) {
return await verifyWebSearchAuth(req, res);
}
const authFields = fieldsMap[toolId];
if (!authFields) {
res.status(404).json({ message: 'Tool not found' });
return;
}
let result;
try {
result = await loadAuthValues({
userId: req.user.id,
authFields,
throwError: false,
});
} catch (error) {
res.status(200).json({ authenticated: false, message: AuthType.USER_PROVIDED });
return;
}
let isUserProvided = false;
for (const field of authFields) {
if (!result[field]) {
res.status(200).json({ authenticated: false, message: AuthType.USER_PROVIDED });
return;
}
if (!isUserProvided && process.env[field] !== result[field]) {
isUserProvided = true;
}
}
res.status(200).json({
authenticated: true,
message: isUserProvided ? AuthType.USER_PROVIDED : AuthType.SYSTEM_DEFINED,
});
} catch (error) {
res.status(500).json({ message: error.message });
}
};
/**
* @param {ServerRequest} req - The request object, containing information about the HTTP request.
* @param {ServerResponse} res - The response object, used to send back the desired HTTP response.
* @param {NextFunction} next - The next middleware function to call.
* @returns {Promise<void>} A promise that resolves when the function has completed.
*/
const callTool = async (req, res) => {
try {
const { toolId = '' } = req.params;
if (!fieldsMap[toolId]) {
logger.warn(`[${toolId}/call] User ${req.user.id} attempted call to invalid tool`);
res.status(404).json({ message: 'Tool not found' });
return;
}
const { partIndex, blockIndex, messageId, conversationId, ...args } = req.body;
if (!messageId) {
logger.warn(`[${toolId}/call] User ${req.user.id} attempted call without message ID`);
res.status(400).json({ message: 'Message ID required' });
return;
}
const message = await getMessage({ user: req.user.id, messageId });
if (!message) {
logger.debug(`[${toolId}/call] User ${req.user.id} attempted call with invalid message ID`);
res.status(404).json({ message: 'Message not found' });
return;
}
logger.debug(`[${toolId}/call] User: ${req.user.id}`);
let hasAccess = true;
if (toolAccessPermType[toolId]) {
hasAccess = await checkAccess(req.user, toolAccessPermType[toolId], [Permissions.USE]);
}
if (!hasAccess) {
logger.warn(
`[${toolAccessPermType[toolId]}] Forbidden: Insufficient permissions for User ${req.user.id}: ${Permissions.USE}`,
);
return res.status(403).json({ message: 'Forbidden: Insufficient permissions' });
}
const { loadedTools } = await loadTools({
user: req.user.id,
tools: [toolId],
functions: true,
options: {
req,
returnMetadata: true,
processFileURL,
uploadImageBuffer,
fileStrategy: req.app.locals.fileStrategy,
},
});
const tool = loadedTools[0];
const toolCallId = `${req.user.id}_${nanoid()}`;
const result = await tool.invoke({
args,
name: toolId,
id: toolCallId,
type: ToolCallTypes.TOOL_CALL,
});
const { content, artifact } = result;
const toolCallData = {
toolId,
messageId,
partIndex,
blockIndex,
conversationId,
result: content,
user: req.user.id,
};
if (!artifact || !artifact.files || toolId !== Tools.execute_code) {
createToolCall(toolCallData).catch((error) => {
logger.error(`Error creating tool call: ${error.message}`);
});
return res.status(200).json({
result: content,
});
}
const artifactPromises = [];
for (const file of artifact.files) {
const { id, name } = file;
artifactPromises.push(
(async () => {
const fileMetadata = await processCodeOutput({
req,
id,
name,
apiKey: tool.apiKey,
messageId,
toolCallId,
conversationId,
session_id: artifact.session_id,
});
if (!fileMetadata) {
return null;
}
return fileMetadata;
})().catch((error) => {
logger.error('Error processing code output:', error);
return null;
}),
);
}
const attachments = await Promise.all(artifactPromises);
toolCallData.attachments = attachments;
createToolCall(toolCallData).catch((error) => {
logger.error(`Error creating tool call: ${error.message}`);
});
res.status(200).json({
result: content,
attachments,
});
} catch (error) {
logger.error('Error calling tool', error);
res.status(500).json({ message: 'Error calling tool' });
}
};
const getToolCalls = async (req, res) => {
try {
const { conversationId } = req.query;
const toolCalls = await getToolCallsByConvo(conversationId, req.user.id);
res.status(200).json(toolCalls);
} catch (error) {
logger.error('Error getting tool calls', error);
res.status(500).json({ message: 'Error getting tool calls' });
}
};
module.exports = {
callTool,
getToolCalls,
verifyToolAuth,
};