@aj-archipelago/cortex
Version:
Cortex is a GraphQL API for AI. It provides a simple, extensible interface for using AI services from OpenAI, Azure and others.
250 lines (238 loc) • 11.6 kB
JavaScript
// sys_tool_cognitive_search.js
// Tool pathway that handles cognitive search across various indexes
import { callPathway } from '../../../../lib/pathwayTools.js';
import { Prompt } from '../../../../server/prompt.js';
import logger from '../../../../lib/logger.js';
import { getSearchResultId } from '../../../../lib/util.js';
export default {
prompt: [],
useInputChunking: false,
enableDuplicateRequests: false,
inputParameters: {
text: '',
filter: '',
top: 50,
titleOnly: false,
stream: false,
indexName: ''
},
timeout: 300,
toolDefinition: [
{
type: "function",
icon: "📂",
function: {
name: "SearchPersonalIndex",
description: "Search through the user's index of personal documents and indexed uploaded files and retrieve the content of the files. Use this tool if the user refers to a file or a document that you don't see uploaded elsewhere in your context. Some file types (e.g. Word documents, Excel documents, very large files, etc.) cannot be attached to a message and will be chunked and indexed and stored in the personal index.",
parameters: {
type: "object",
properties: {
text: {
type: "string",
description: "The search query to find relevant content in personal documents. Can be a specific phrase or '*' for all documents, or a query formatted with AI Search syntax."
},
filter: {
type: "string",
description: "Optional OData filter expression for date filtering (e.g. 'date ge 2024-02-22T00:00:00Z')"
},
top: {
type: "integer",
description: "Number of results to return (default is 50)"
},
titleOnly: {
type: "boolean",
description: "If true, only return document titles without content - faster and great for counting results"
},
userMessage: {
type: "string",
description: "A user-friendly message that describes what you're doing with this tool"
}
},
required: ["text", "userMessage"]
}
}
},
{
type: "function",
icon: "📰",
function: {
name: "SearchAJA",
description: "Search Al Jazeera Arabic news articles. Use this for finding Arabic news content including the latest news and articles. Make sure to include a date filter when looking for recent articles.",
parameters: {
type: "object",
properties: {
text: {
type: "string",
description: "The search query in Arabic to find relevant news articles. Can be a specific phrase or '*' for all articles, or a query formatted with AI Search syntax."
},
filter: {
type: "string",
description: "Optional OData filter expression for date filtering (e.g. 'date ge 2024-02-22T00:00:00Z')"
},
top: {
type: "integer",
description: "Number of results to return (default is 50)"
},
titleOnly: {
type: "boolean",
description: "If true, only return article titles without content - faster and great for counting results"
},
userMessage: {
type: "string",
description: "A user-friendly message that describes what you're doing with this tool"
}
},
required: ["text", "userMessage"]
}
}
},
{
type: "function",
icon: "📰",
function: {
name: "SearchAJE",
description: "Search Al Jazeera English news articles. Use this for finding English news content including the latest news and articles. Make sure to include a date filter when looking for recent articles.",
parameters: {
type: "object",
properties: {
text: {
type: "string",
description: "The search query in English to find relevant news articles. Can be a specific phrase or '*' for all articles, or a query formatted with AI Search syntax."
},
filter: {
type: "string",
description: "Optional OData filter expression for date filtering (e.g. 'date ge 2024-02-22T00:00:00Z')"
},
top: {
type: "integer",
description: "Number of results to return (default is 50)"
},
titleOnly: {
type: "boolean",
description: "If true, only return article titles without content - faster and great for counting results"
},
userMessage: {
type: "string",
description: "A user-friendly message that describes what you're doing with this tool"
}
},
required: ["text", "userMessage"]
}
}
},
{
type: "function",
icon: "⚡️",
function: {
name: "SearchWires",
description: "Search news wires from Reuters, AFP, AP, and other news agencies. Use this for finding the latest news and articles from the wires. Make sure to include a date filter when looking for recent articles.",
parameters: {
type: "object",
properties: {
text: {
type: "string",
description: "The search query to find relevant news wires. Can be a specific phrase or '*' for all wires, or a query formatted with AI Search syntax."
},
filter: {
type: "string",
description: "Optional OData filter expression for date filtering (e.g. 'date ge 2024-02-22T00:00:00Z')"
},
top: {
type: "integer",
description: "Number of results to return (default is 50)"
},
titleOnly: {
type: "boolean",
description: "If true, only return wire titles without content - faster and great for counting results"
},
userMessage: {
type: "string",
description: "A user-friendly message that describes what you're doing with this tool"
}
},
required: ["text", "userMessage"]
}
}
}
],
executePathway: async ({args, runAllPrompts, resolver}) => {
const { text, filter, top, titleOnly, stream, chatId, indexName, semanticConfiguration } = args;
// Map tool names to index names
const toolToIndex = {
'searchpersonalindex': 'indexcortex',
'searchaja': 'indexucmsaja',
'searchaje': 'indexucmsaje',
'searchwires': 'indexwires'
};
// Helper function to remove vector fields from search results
const removeVectorFields = (result) => {
const { text_vector, image_vector, ...cleanResult } = result;
return cleanResult;
};
// Get the tool name from the function call
const toolName = args.toolFunction;
const toolIndexName = indexName || toolToIndex[toolName];
if (!toolName || !toolIndexName) {
throw new Error(`Invalid tool name: ${toolName}. Search not allowed.`);
}
try {
// Call the cognitive search pathway
const response = await callPathway('cognitive_search', {
...args,
text,
filter,
top: top || 50,
titleOnly: titleOnly || false,
indexName: toolIndexName,
semanticConfiguration,
stream: stream || false,
chatId
});
const parsedResponse = JSON.parse(response);
const combinedResults = [];
// Add OData context and count information if present
if (parsedResponse["@odata.context"]) {
combinedResults.push({
searchResultId: getSearchResultId(),
key: "@odata.context",
content: parsedResponse["@odata.context"],
source_type: 'metadata'
});
}
if (parsedResponse["@odata.count"]) {
combinedResults.push({
searchResultId: getSearchResultId(),
key: "@odata.count",
content: parsedResponse["@odata.count"].toString(),
source_type: 'metadata'
});
}
if (parsedResponse.value && Array.isArray(parsedResponse.value)) {
// Filter out vector fields from each result before adding to combinedResults
combinedResults.push(...parsedResponse.value.map(result => ({
...removeVectorFields(result),
searchResultId: getSearchResultId()
})));
}
// Extract semantic answers
const answers = parsedResponse["@search.answers"];
if (answers && Array.isArray(answers)) {
const formattedAnswers = answers.map(ans => ({
// Create a pseudo-document structure for answers
searchResultId: getSearchResultId(),
title: "", // no title for answers
content: ans.text || "", // Use text as content
key: ans.key, // Keep the key if needed later
score: ans.score, // Keep score
source_type: 'answer' // Add a type identifier
// url: null - Answers don't have URLs
}));
combinedResults.push(...formattedAnswers);
}
return JSON.stringify({ _type: "SearchResponse", value: combinedResults });
} catch (e) {
logger.error(`Error in cognitive search for index ${indexName}: ${e}`);
throw e;
}
}
};