mirror of
https://github.com/danny-avila/LibreChat.git
synced 2025-12-19 01:40:15 +01:00
* 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
723 lines
22 KiB
JavaScript
723 lines
22 KiB
JavaScript
const fs = require('fs');
|
|
const path = require('path');
|
|
const { zodToJsonSchema } = require('zod-to-json-schema');
|
|
const { Calculator } = require('@langchain/community/tools/calculator');
|
|
const { tool: toolFn, Tool, DynamicStructuredTool } = require('@langchain/core/tools');
|
|
const {
|
|
Tools,
|
|
ErrorTypes,
|
|
ContentTypes,
|
|
imageGenTools,
|
|
EToolResources,
|
|
EModelEndpoint,
|
|
actionDelimiter,
|
|
ImageVisionTool,
|
|
openapiToFunction,
|
|
AgentCapabilities,
|
|
validateAndParseOpenAPISpec,
|
|
} = require('librechat-data-provider');
|
|
const {
|
|
createActionTool,
|
|
decryptMetadata,
|
|
loadActionSets,
|
|
domainParser,
|
|
} = require('./ActionService');
|
|
const {
|
|
createOpenAIImageTools,
|
|
createYouTubeTools,
|
|
manifestToolMap,
|
|
toolkits,
|
|
} = require('~/app/clients/tools');
|
|
const { processFileURL, uploadImageBuffer } = require('~/server/services/Files/process');
|
|
const { createOnSearchResults } = require('~/server/services/Tools/search');
|
|
const { isActionDomainAllowed } = require('~/server/services/domains');
|
|
const { getEndpointsConfig } = require('~/server/services/Config');
|
|
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');
|
|
|
|
/**
|
|
* @param {string} toolName
|
|
* @returns {string | undefined} toolKey
|
|
*/
|
|
function getToolkitKey(toolName) {
|
|
/** @type {string|undefined} */
|
|
let toolkitKey;
|
|
for (const toolkit of toolkits) {
|
|
if (toolName.startsWith(EToolResources.image_edit)) {
|
|
const splitMatches = toolkit.pluginKey.split('_');
|
|
const suffix = splitMatches[splitMatches.length - 1];
|
|
if (toolName.endsWith(suffix)) {
|
|
toolkitKey = toolkit.pluginKey;
|
|
break;
|
|
}
|
|
}
|
|
if (toolName.startsWith(toolkit.pluginKey)) {
|
|
toolkitKey = toolkit.pluginKey;
|
|
break;
|
|
}
|
|
}
|
|
return toolkitKey;
|
|
}
|
|
|
|
/**
|
|
* 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]);
|
|
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 & Toolkits; schema: { input: string } */
|
|
const basicToolInstances = [
|
|
new Calculator(),
|
|
...createOpenAIImageTools({ override: true }),
|
|
...createYouTubeTools({ override: true }),
|
|
];
|
|
for (const toolInstance of basicToolInstances) {
|
|
const formattedTool = formatToOpenAIAssistantTool(toolInstance);
|
|
let toolName = formattedTool[Tools.function].name;
|
|
toolName = getToolkitKey(toolName) ?? toolName;
|
|
if (filter.has(toolName) && included.size === 0) {
|
|
continue;
|
|
}
|
|
|
|
if (included.size > 0 && !included.has(toolName)) {
|
|
continue;
|
|
}
|
|
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 && 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: ${requiredActions[0].run_id}`,
|
|
requiredActions,
|
|
);
|
|
const toolDefinitions = client.req.app.locals.availableTools;
|
|
const seenToolkits = new Set();
|
|
const tools = requiredActions
|
|
.map((action) => {
|
|
const toolName = action.tool;
|
|
const toolDef = toolDefinitions[toolName];
|
|
if (toolDef && !manifestToolMap[toolName]) {
|
|
for (const toolkit of toolkits) {
|
|
if (seenToolkits.has(toolkit.pluginKey)) {
|
|
return;
|
|
} else if (toolName.startsWith(`${toolkit.pluginKey}_`)) {
|
|
seenToolkits.add(toolkit.pluginKey);
|
|
return toolkit.pluginKey;
|
|
}
|
|
}
|
|
}
|
|
return toolName;
|
|
})
|
|
.filter((toolName) => !!toolName);
|
|
|
|
const { loadedTools } = await loadTools({
|
|
user: client.req.user.id,
|
|
model: client.req.body.model ?? 'gpt-4o-mini',
|
|
tools,
|
|
functions: true,
|
|
endpoint: client.req.body.endpoint,
|
|
options: {
|
|
processFileURL,
|
|
req: client.req,
|
|
uploadImageBuffer,
|
|
openAIApiKey: client.apiKey,
|
|
fileStrategy: client.req.app.locals.fileStrategy,
|
|
returnMetadata: 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.`);
|
|
|
|
// Load all action sets once if not already loaded
|
|
if (!actionSets.length) {
|
|
actionSets =
|
|
(await loadActionSets({
|
|
assistant_id: client.req.body.assistant_id,
|
|
})) ?? [];
|
|
|
|
// Process all action sets once
|
|
// Map domains to their processed action sets
|
|
const processedDomains = new Map();
|
|
const domainMap = new Map();
|
|
|
|
for (const action of actionSets) {
|
|
const domain = await domainParser(action.metadata.domain, true);
|
|
domainMap.set(domain, action);
|
|
|
|
// Check if domain is allowed
|
|
const isDomainAllowed = await isActionDomainAllowed(action.metadata.domain);
|
|
if (!isDomainAllowed) {
|
|
continue;
|
|
}
|
|
|
|
// Validate and parse OpenAPI spec
|
|
const validationResult = validateAndParseOpenAPISpec(action.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}`,
|
|
);
|
|
}
|
|
|
|
// Process the OpenAPI spec
|
|
const { requestBuilders } = openapiToFunction(validationResult.spec);
|
|
|
|
// Store encrypted values for OAuth flow
|
|
const encrypted = {
|
|
oauth_client_id: action.metadata.oauth_client_id,
|
|
oauth_client_secret: action.metadata.oauth_client_secret,
|
|
};
|
|
|
|
// Decrypt metadata
|
|
const decryptedAction = { ...action };
|
|
decryptedAction.metadata = await decryptMetadata(action.metadata);
|
|
|
|
processedDomains.set(domain, {
|
|
action: decryptedAction,
|
|
requestBuilders,
|
|
encrypted,
|
|
});
|
|
|
|
// Store builders for reuse
|
|
ActionBuildersMap[action.metadata.domain] = requestBuilders;
|
|
}
|
|
|
|
// Update actionSets reference to use the domain map
|
|
actionSets = { domainMap, processedDomains };
|
|
}
|
|
|
|
// Find the matching domain for this tool
|
|
let currentDomain = '';
|
|
for (const domain of actionSets.domainMap.keys()) {
|
|
if (currentAction.tool.includes(domain)) {
|
|
currentDomain = domain;
|
|
break;
|
|
}
|
|
}
|
|
|
|
if (!currentDomain || !actionSets.processedDomains.has(currentDomain)) {
|
|
// TODO: try `function` if no action set is found
|
|
// throw new Error(`Tool ${currentAction.tool} not found.`);
|
|
continue;
|
|
}
|
|
|
|
const { action, requestBuilders, encrypted } = actionSets.processedDomains.get(currentDomain);
|
|
const functionName = currentAction.tool.replace(`${actionDelimiter}${currentDomain}`, '');
|
|
const requestBuilder = requestBuilders[functionName];
|
|
|
|
if (!requestBuilder) {
|
|
// throw new Error(`Tool ${currentAction.tool} not found.`);
|
|
continue;
|
|
}
|
|
|
|
// We've already decrypted the metadata, so we can pass it directly
|
|
tool = await createActionTool({
|
|
userId: client.req.user.id,
|
|
res: client.res,
|
|
action,
|
|
requestBuilder,
|
|
// Note: intentionally not passing zodSchema, name, and description for assistants API
|
|
encrypted, // Pass the encrypted values for OAuth flow
|
|
});
|
|
if (!tool) {
|
|
logger.warn(
|
|
`Invalid action: user: ${client.req.user.id} | thread_id: ${requiredActions[0].thread_id} | run_id: ${requiredActions[0].run_id} | toolName: ${currentAction.tool}`,
|
|
);
|
|
throw new Error(`{"type":"${ErrorTypes.INVALID_ACTION}"}`);
|
|
}
|
|
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 the tool classes.
|
|
* @param {Object} params - Run params containing user and request information.
|
|
* @param {ServerRequest} params.req - The request object.
|
|
* @param {ServerResponse} params.res - The request object.
|
|
* @param {Pick<Agent, 'id' | 'provider' | 'model' | 'tools'} params.agent - The agent to load tools for.
|
|
* @param {string | undefined} [params.openAIApiKey] - The OpenAI API key.
|
|
* @returns {Promise<{ tools?: StructuredTool[] }>} The agent tools.
|
|
*/
|
|
async function loadAgentTools({ req, res, agent, tool_resources, openAIApiKey }) {
|
|
if (!agent.tools || agent.tools.length === 0) {
|
|
return {};
|
|
}
|
|
|
|
const endpointsConfig = await getEndpointsConfig(req);
|
|
const enabledCapabilities = new Set(endpointsConfig?.[EModelEndpoint.agents]?.capabilities ?? []);
|
|
const checkCapability = (capability) => enabledCapabilities.has(capability);
|
|
const areToolsEnabled = checkCapability(AgentCapabilities.tools);
|
|
|
|
let includesWebSearch = false;
|
|
const _agentTools = agent.tools?.filter((tool) => {
|
|
if (tool === Tools.file_search) {
|
|
return checkCapability(AgentCapabilities.file_search);
|
|
} else if (tool === Tools.execute_code) {
|
|
return checkCapability(AgentCapabilities.execute_code);
|
|
} else if (tool === Tools.web_search) {
|
|
includesWebSearch = checkCapability(AgentCapabilities.web_search);
|
|
return includesWebSearch;
|
|
} else if (!areToolsEnabled && !tool.includes(actionDelimiter)) {
|
|
return false;
|
|
}
|
|
return true;
|
|
});
|
|
|
|
if (!_agentTools || _agentTools.length === 0) {
|
|
return {};
|
|
}
|
|
/** @type {ReturnType<createOnSearchResults>} */
|
|
let webSearchCallbacks;
|
|
if (includesWebSearch) {
|
|
webSearchCallbacks = createOnSearchResults(res);
|
|
}
|
|
const { loadedTools, toolContextMap } = await loadTools({
|
|
agent,
|
|
functions: true,
|
|
user: req.user.id,
|
|
tools: _agentTools,
|
|
options: {
|
|
req,
|
|
openAIApiKey,
|
|
tool_resources,
|
|
processFileURL,
|
|
uploadImageBuffer,
|
|
returnMetadata: true,
|
|
fileStrategy: req.app.locals.fileStrategy,
|
|
[Tools.web_search]: webSearchCallbacks,
|
|
},
|
|
});
|
|
|
|
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;
|
|
}
|
|
|
|
if (!areToolsEnabled) {
|
|
continue;
|
|
}
|
|
|
|
if (tool.mcp === true) {
|
|
agentTools.push(tool);
|
|
continue;
|
|
}
|
|
|
|
if (tool instanceof DynamicStructuredTool) {
|
|
agentTools.push(tool);
|
|
continue;
|
|
}
|
|
|
|
const toolDefinition = {
|
|
name: tool.name,
|
|
schema: tool.schema,
|
|
description: tool.description,
|
|
};
|
|
|
|
if (imageGenTools.has(tool.name)) {
|
|
toolDefinition.responseFormat = 'content_and_artifact';
|
|
}
|
|
|
|
const toolInstance = toolFn(async (...args) => {
|
|
return tool['_call'](...args);
|
|
}, toolDefinition);
|
|
|
|
agentTools.push(toolInstance);
|
|
}
|
|
|
|
const ToolMap = loadedTools.reduce((map, tool) => {
|
|
map[tool.name] = tool;
|
|
return map;
|
|
}, {});
|
|
|
|
if (!checkCapability(AgentCapabilities.actions)) {
|
|
return {
|
|
tools: agentTools,
|
|
toolContextMap,
|
|
};
|
|
}
|
|
|
|
const actionSets = (await loadActionSets({ agent_id: agent.id })) ?? [];
|
|
if (actionSets.length === 0) {
|
|
if (_agentTools.length > 0 && agentTools.length === 0) {
|
|
logger.warn(`No tools found for the specified tool calls: ${_agentTools.join(', ')}`);
|
|
}
|
|
return {
|
|
tools: agentTools,
|
|
toolContextMap,
|
|
};
|
|
}
|
|
|
|
// Process each action set once (validate spec, decrypt metadata)
|
|
const processedActionSets = new Map();
|
|
const domainMap = new Map();
|
|
|
|
for (const action of actionSets) {
|
|
const domain = await domainParser(action.metadata.domain, true);
|
|
domainMap.set(domain, action);
|
|
|
|
// Check if domain is allowed (do this once per action set)
|
|
const isDomainAllowed = await isActionDomainAllowed(action.metadata.domain);
|
|
if (!isDomainAllowed) {
|
|
continue;
|
|
}
|
|
|
|
// Validate and parse OpenAPI spec once per action set
|
|
const validationResult = validateAndParseOpenAPISpec(action.metadata.raw_spec);
|
|
if (!validationResult.spec) {
|
|
continue;
|
|
}
|
|
|
|
const encrypted = {
|
|
oauth_client_id: action.metadata.oauth_client_id,
|
|
oauth_client_secret: action.metadata.oauth_client_secret,
|
|
};
|
|
|
|
// Decrypt metadata once per action set
|
|
const decryptedAction = { ...action };
|
|
decryptedAction.metadata = await decryptMetadata(action.metadata);
|
|
|
|
// Process the OpenAPI spec once per action set
|
|
const { requestBuilders, functionSignatures, zodSchemas } = openapiToFunction(
|
|
validationResult.spec,
|
|
true,
|
|
);
|
|
|
|
processedActionSets.set(domain, {
|
|
action: decryptedAction,
|
|
requestBuilders,
|
|
functionSignatures,
|
|
zodSchemas,
|
|
encrypted,
|
|
});
|
|
}
|
|
|
|
// Now map tools to the processed action sets
|
|
const ActionToolMap = {};
|
|
|
|
for (const toolName of _agentTools) {
|
|
if (ToolMap[toolName]) {
|
|
continue;
|
|
}
|
|
|
|
// Find the matching domain for this tool
|
|
let currentDomain = '';
|
|
for (const domain of domainMap.keys()) {
|
|
if (toolName.includes(domain)) {
|
|
currentDomain = domain;
|
|
break;
|
|
}
|
|
}
|
|
|
|
if (!currentDomain || !processedActionSets.has(currentDomain)) {
|
|
continue;
|
|
}
|
|
|
|
const { action, encrypted, zodSchemas, requestBuilders, functionSignatures } =
|
|
processedActionSets.get(currentDomain);
|
|
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({
|
|
userId: req.user.id,
|
|
res,
|
|
action,
|
|
requestBuilder,
|
|
zodSchema,
|
|
encrypted,
|
|
name: toolName,
|
|
description: functionSig.description,
|
|
});
|
|
|
|
if (!tool) {
|
|
logger.warn(
|
|
`Invalid action: user: ${req.user.id} | agent_id: ${agent.id} | toolName: ${toolName}`,
|
|
);
|
|
throw new Error(`{"type":"${ErrorTypes.INVALID_ACTION}"}`);
|
|
}
|
|
|
|
agentTools.push(tool);
|
|
ActionToolMap[toolName] = tool;
|
|
}
|
|
}
|
|
|
|
if (_agentTools.length > 0 && agentTools.length === 0) {
|
|
logger.warn(`No tools found for the specified tool calls: ${_agentTools.join(', ')}`);
|
|
return {};
|
|
}
|
|
|
|
return {
|
|
tools: agentTools,
|
|
toolContextMap,
|
|
};
|
|
}
|
|
|
|
module.exports = {
|
|
getToolkitKey,
|
|
loadAgentTools,
|
|
loadAndFormatTools,
|
|
processRequiredActions,
|
|
formatToOpenAIAssistantTool,
|
|
};
|