🪶 feat: Add Support for Azure OpenAI Base URL (#1596)

* refactor(extractBaseURL): add handling for all possible Cloudflare AI Gateway endpoints

* chore: added endpointoption todo for updating type and optimizing handling app-wide

* feat(azureUtils):
- `genAzureChatCompletion`: allow optional client pass to update azure property
- `constructAzureURL`: optionally replace placeholders for instance and deployment names of an azure baseURL
- add tests for module

* refactor(extractBaseURL): return entire input when cloudflare `azure-openai` suffix detected
- also add more tests for both construct and extract URL

* refactor(genAzureChatCompletion): only allow omitting instance name if baseURL is not set

* refactor(initializeClient): determine `reverseProxyUrl` based on endpoint (azure or openai)

* refactor: utitlize `constructAzureURL` when `AZURE_OPENAI_BASEURL` is set

* docs: update docs on `AZURE_OPENAI_BASEURL`

* fix(ci): update expected error message for `azureUtils` tests

api/app/clients/OpenAIClient.js

@@ -2,8 +2,13 @@ const OpenAI = require('openai');
 const { HttpsProxyAgent } = require('https-proxy-agent');
 const { getResponseSender, ImageDetailCost, ImageDetail } = require('librechat-data-provider');
 const { encoding_for_model: encodingForModel, get_encoding: getEncoding } = require('tiktoken');
+const {
+  getModelMaxTokens,
+  genAzureChatCompletion,
+  extractBaseURL,
+  constructAzureURL,
+} = require('~/utils');
 const { encodeAndFormat, validateVisionModel } = require('~/server/services/Files/images');
-const { getModelMaxTokens, genAzureChatCompletion, extractBaseURL } = require('~/utils');
 const { truncateText, formatMessage, CUT_OFF_PROMPT } = require('./prompts');
 const { handleOpenAIErrors } = require('./tools/util');
 const spendTokens = require('~/models/spendTokens');
@@ -32,6 +37,7 @@ class OpenAIClient extends BaseClient {
       ? options.contextStrategy.toLowerCase()
       : 'discard';
     this.shouldSummarize = this.contextStrategy === 'summarize';
+    /** @type {AzureOptions} */
     this.azure = options.azure || false;
     this.setOptions(options);
   }
@@ -104,10 +110,10 @@ class OpenAIClient extends BaseClient {
     }
 
     if (this.azure && process.env.AZURE_OPENAI_DEFAULT_MODEL) {
-      this.azureEndpoint = genAzureChatCompletion(this.azure, this.modelOptions.model);
+      this.azureEndpoint = genAzureChatCompletion(this.azure, this.modelOptions.model, this);
       this.modelOptions.model = process.env.AZURE_OPENAI_DEFAULT_MODEL;
     } else if (this.azure) {
-      this.azureEndpoint = genAzureChatCompletion(this.azure, this.modelOptions.model);
+      this.azureEndpoint = genAzureChatCompletion(this.azure, this.modelOptions.model, this);
     }
 
     const { model } = this.modelOptions;
@@ -711,7 +717,7 @@ class OpenAIClient extends BaseClient {
 
       if (this.azure) {
         modelOptions.model = process.env.AZURE_OPENAI_DEFAULT_MODEL ?? modelOptions.model;
-        this.azureEndpoint = genAzureChatCompletion(this.azure, modelOptions.model);
+        this.azureEndpoint = genAzureChatCompletion(this.azure, modelOptions.model, this);
       }
 
       const instructionsPayload = [
@@ -949,7 +955,12 @@ ${convo}
         // Azure does not accept `model` in the body, so we need to remove it.
         delete modelOptions.model;
 
-        opts.baseURL = this.azureEndpoint.split('/chat')[0];
+        opts.baseURL = this.langchainProxy
+          ? constructAzureURL({
+            baseURL: this.langchainProxy,
+            azure: this.azure,
+          })
+          : this.azureEndpoint.split(/\/(chat|completion)/)[0];
         opts.defaultQuery = { 'api-version': this.azure.azureOpenAIApiVersion };
         opts.defaultHeaders = { ...opts.defaultHeaders, 'api-key': this.apiKey };
       }

api/app/clients/llm/createLLM.js

@@ -1,6 +1,6 @@
 const { ChatOpenAI } = require('langchain/chat_models/openai');
-const { sanitizeModelName } = require('../../../utils');
-const { isEnabled } = require('../../../server/utils');
+const { sanitizeModelName, constructAzureURL } = require('~/utils');
+const { isEnabled } = require('~/server/utils');
 
 /**
  * Creates a new instance of a language model (LLM) for chat interactions.
@@ -36,6 +36,7 @@ function createLLM({
     apiKey: openAIApiKey,
   };
 
+  /**  @type {AzureOptions} */
   let azureOptions = {};
   if (azure) {
     const useModelName = isEnabled(process.env.AZURE_USE_MODEL_AS_DEPLOYMENT_NAME);
@@ -53,8 +54,12 @@ function createLLM({
     modelOptions.modelName = process.env.AZURE_OPENAI_DEFAULT_MODEL;
   }
 
-  // console.debug('createLLM: configOptions');
-  // console.debug(configOptions);
+  if (azure && configOptions.basePath) {
+    configOptions.basePath = constructAzureURL({
+      baseURL: configOptions.basePath,
+      azure: azureOptions,
+    });
+  }
 
   return new ChatOpenAI(
     {