@aj-archipelago/cortex 1.1.6 → 1.1.7

package/config.js

@@ -146,6 +146,16 @@ var config = convict({
                 "maxTokenLength": 128000,
                 "supportsStreaming": true
             },
+            "azure-bing": {
+                "type": "AZURE-BING",
+                "url": "https://api.bing.microsoft.com/v7.0/search",
+                "headers": {
+                    "Ocp-Apim-Subscription-Key": "{{AZURE_BING_KEY}}",
+                    "Content-Type": "application/json"
+                },
+                "requestsPerSecond": 10,
+                "maxTokenLength": 200000
+            },
         },
         env: 'CORTEX_MODELS'
     },

package/lib/cortexRequest.js

@@ -11,6 +11,7 @@ class CortexRequest {
         this._pathwayResolver = pathwayResolver || {};
         this._selectedEndpoint = selectedEndpoint || {};
         this._stream = stream || false;
+        this._method = 'POST';
 
         if (this._pathwayResolver) {
             this._model = this._pathwayResolver.model;
@@ -41,6 +42,15 @@ class CortexRequest {
         this._url = value;
     }
 
+    // method getter and setter
+    get method() {
+        return this._method;
+    }
+
+    set method(value) {
+        this._method = value;
+    }
+
     // data getter and setter
     get data() {
         return this._data;

package/lib/requestExecutor.js

@@ -183,11 +183,15 @@ setInterval(() => {
   }
 }, 30000); // Log rates every 30 seconds
 
-const postWithMonitor = async (endpoint, url, data, axiosConfigObj) => {
+const requestWithMonitor = async (endpoint, url, data, axiosConfigObj) => {
     const callId = endpoint?.monitor?.startCall();
     let response;
     try {
-        response = await cortexAxios.post(url, data, axiosConfigObj);
+        if (axiosConfigObj?.method == 'GET'){
+            response = await cortexAxios.get(url, axiosConfigObj);
+        } else {
+            response = await cortexAxios.post(url, data, axiosConfigObj);
+        }
     } catch (error) {
         // throw new error with duration as part of the error data
         throw { ...error, duration: endpoint?.monitor?.incrementErrorCount(callId, error?.response?.status || null) };
@@ -213,22 +217,22 @@ const getDuplicateRequestDelay = (index, duplicateRequestAfter) => {
     return duplicateRequestTimeout;
 }
 
-const postRequest = async (cortexRequest) => {
+const makeRequest = async (cortexRequest) => {
     let promises = [];
     // retry certain errors up to MAX_RETRY times
     for (let i = 0; i < MAX_RETRY; i++) {
-        const { url, data, params, headers, cache, selectedEndpoint, requestId, pathway, model, stream} = cortexRequest;
+        const { url, data, params, headers, cache, selectedEndpoint, requestId, pathway, model, stream, method} = cortexRequest;
         const enableDuplicateRequests = pathway?.enableDuplicateRequests !== undefined ? pathway.enableDuplicateRequests : config.get('enableDuplicateRequests');
         const maxDuplicateRequests = enableDuplicateRequests ? MAX_DUPLICATE_REQUESTS : 1;
         const duplicateRequestAfter = (pathway?.duplicateRequestAfter || DUPLICATE_REQUEST_AFTER) * 1000;
 
-        const axiosConfigObj = { params, headers, cache };
+        const axiosConfigObj = { params, headers, cache, method };
         const streamRequested = (stream || params?.stream || data?.stream);
         // if we're using streaming, duplicate requests are
         // not supported, so we just push one promise into the array
         if (streamRequested && model.supportsStreaming) {
             axiosConfigObj.responseType = 'stream';
-            promises.push(selectedEndpoint.limiter.schedule({expiration: pathway.timeout * 1000 + 1000, id: `${requestId}_${uuidv4()}`},() => postWithMonitor(selectedEndpoint, url, data, axiosConfigObj)));
+            promises.push(selectedEndpoint.limiter.schedule({expiration: pathway.timeout * 1000 + 1000, id: `${requestId}_${uuidv4()}`},() => requestWithMonitor(selectedEndpoint, url, data, axiosConfigObj)));
         } else {
             if (streamRequested) {
                 logger.info(`>>> [${requestId}] ${model} does not support streaming - sending non-streaming request`);
@@ -253,7 +257,7 @@ const postRequest = async (cortexRequest) => {
                             if (!selectedEndpoint.limiter) {
                                 throw new Error(`No limiter for endpoint ${endpointName}!`);
                             }
-                            const axiosConfigObj = { params, headers, cache };
+                            const axiosConfigObj = { params, headers, cache, method };
 
                             let response = null;
                             let duration = null;
@@ -269,7 +273,7 @@ const postRequest = async (cortexRequest) => {
                                     logger.info(`\n${header}\n${logMessage}`);
                                 }
 
-                                ({ response, duration } = await selectedEndpoint.limiter.schedule({expiration: pathway.timeout * 1000 + 1000, id: `${requestId}_${uuidv4()}`}, () => postWithMonitor(selectedEndpoint, url, data, axiosConfigObj)));
+                                ({ response, duration } = await selectedEndpoint.limiter.schedule({expiration: pathway.timeout * 1000 + 1000, id: `${requestId}_${uuidv4()}`}, () => requestWithMonitor(selectedEndpoint, url, data, axiosConfigObj)));
 
                                 if (!controller.signal?.aborted) {
                                     logger.debug(`<<< [${requestId}] received response for request ${index}`);
@@ -341,7 +345,7 @@ const postRequest = async (cortexRequest) => {
 
 const executeRequest = async (cortexRequest) => {
     try {
-        const { response, duration } = await postRequest(cortexRequest);
+        const { response, duration } = await makeRequest(cortexRequest);
         const requestId = cortexRequest.requestId;
         const { error, data, cached } = response;
         if (cached) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@aj-archipelago/cortex",
-  "version": "1.1.6",
+  "version": "1.1.7",
   "description": "Cortex is a GraphQL API for AI. It provides a simple, extensible interface for using AI services from OpenAI, Azure and others.",
   "private": false,
   "repository": {
@@ -52,7 +52,7 @@
     "handlebars": "^4.7.7",
     "ioredis": "^5.3.1",
     "keyv": "^4.5.2",
-    "langchain": "^0.0.47",
+    "langchain": "^0.1.28",
     "mime-types": "^2.1.35",
     "subsrt": "^1.1.1",
     "uuid": "^9.0.0",

package/pathways/bing.js

@@ -0,0 +1,12 @@
+// bing.js
+// Web search tool
+
+export default {
+    inputParameters: {
+        text: ``,
+    },
+    timeout: 400,
+    enableDuplicateRequests: false,
+    model: 'azure-bing',
+};
+

package/pathways/index.js

@@ -19,8 +19,10 @@ import transcribe from './transcribe.js';
 import translate from './translate.js';
 import embeddings from './embeddings.js';
 import vision from './vision.js';
+import bing from './bing.js';
 
 export {
+    bing,
     edit,
     chat,
     bias,

package/server/modelExecutor.js

@@ -19,6 +19,7 @@ import OpenAIDallE3Plugin from './plugins/openAiDallE3Plugin.js';
 import OpenAIVisionPlugin from './plugins/openAiVisionPlugin.js';
 import GeminiChatPlugin from './plugins/geminiChatPlugin.js';
 import GeminiVisionPlugin from './plugins/geminiVisionPlugin.js';
+import AzureBingPlugin from './plugins/azureBingPlugin.js';
 
 class ModelExecutor {
     constructor(pathway, model) {
@@ -80,6 +81,9 @@ class ModelExecutor {
             case 'GEMINI-VISION':
                 plugin = new GeminiVisionPlugin(pathway, model);
                 break;
+            case 'AZURE-BING':
+                plugin = new AzureBingPlugin(pathway, model);
+                break;
             default:
                 throw new Error(`Unsupported model type: ${model.type}`);
         }

package/server/plugins/azureBingPlugin.js

@@ -0,0 +1,44 @@
+import ModelPlugin from './modelPlugin.js';
+import logger from '../../lib/logger.js';
+
+class AzureBingPlugin extends ModelPlugin {
+    constructor(pathway, model) {
+        super(pathway, model);
+    }
+    
+    getRequestParameters(text) {
+        const requestParameters = {
+            data: [
+            ],
+            params: {
+                q: text,
+            }
+        };
+        return requestParameters;
+    }
+
+    async execute(text, parameters, prompt, cortexRequest) {
+        const requestParameters = this.getRequestParameters(text, parameters, prompt);
+
+        cortexRequest.data = requestParameters.data;
+        cortexRequest.params = requestParameters.params;
+        cortexRequest.method = 'GET';
+
+        return this.executeRequest(cortexRequest);
+    }
+    
+    parseResponse(data) {
+        return JSON.stringify(data);
+    }
+    
+    // Override the logging function to display the request and response
+    logRequestData(data, responseData, prompt) {
+        this.logAIRequestFinished();
+    
+        logger.debug(`${this.parseResponse(responseData)}`);
+    
+        prompt && prompt.debugInfo && (prompt.debugInfo += `\n${JSON.stringify(data)}`);
+    }
+}
+
+export default AzureBingPlugin;