Merge e4c3661c91 into 64ee19c70a

package.json

@@ -29,6 +29,7 @@
   "dependencies": {
     "@iarna/toml": "^2.2.5",
     "@langchain/anthropic": "^0.2.3",
+    "@langchain/azure-openai": "^0.0.4",
     "@langchain/community": "^0.2.16",
     "@langchain/openai": "^0.0.25",
     "@xenova/transformers": "^2.17.1",

sample.config.toml

@@ -7,7 +7,21 @@ KEEP_ALIVE = "5m" # How long to keep Ollama models loaded into memory. (Instead
 OPENAI = "" # OpenAI API key - sk-1234567890abcdef1234567890abcdef
 GROQ = "" # Groq API key - gsk_1234567890abcdef1234567890abcdef
 ANTHROPIC = "" # Anthropic API key - sk-ant-1234567890abcdef1234567890abcdef
+AZURE = "" # Azure OpenAI API key
 
 [API_ENDPOINTS]
 SEARXNG = "http://localhost:32768" # SearxNG API URL
-OLLAMA = "" # Ollama API URL - http://host.docker.internal:11434
+OLLAMA = "" # Ollama API URL - http://host.docker.internal:11434
+AZURE = "" # Azure OpenAI Endpoint URL - https://endpointhere.openai.azure.com/
+
+[AZURE_OPENAI]
+DEPLOYMENT_NAME_GPT35_TURBO = "gpt-35-turbo" # Deployment name for GPT-3.5 turbo
+DEPLOYMENT_NAME_GPT35_TURBO_16K = "gpt-35-turbo-16k" # Deployment name for GPT-3.5-turbo 16k
+DEPLOYMENT_NAME_GPT35_TURBO_INSTRUCT = "gpt-35-turbo-instruct" # Deployment name for GPT-3.5 turbo instruct
+DEPLOYMENT_NAME_GPT4 = "gpt4" # Deployment name for GPT-4
+DEPLOYMENT_NAME_GPT4_32K = "gpt432k" # Deployment name for GPT-4 32K
+DEPLOYMENT_NAME_GPT4_OMNI = "gpt-4o" # Deployment name for GPT-4 omni
+DEPLOYMENT_NAME_GPT4_OMNI_MINI = "gpt-4o-mini" # Deployment name for GPT-4 omni mini
+DEPLOYMENT_NAME_EMBEDDINGS_ADA = "text-embeddings-ada-002" # Deployment name for text embeddings ada 002
+DEPLOYMENT_NAME_EMBEDDINGS_SMALL = "text-embedding-3-small" # Deployment name for text embedding 3 small
+DEPLOYMENT_NAME_EMBEDDINGS_LARGE = "text-embedding-3-large" # Deployment name for text embedding 3 large

src/config.ts

@@ -14,10 +14,24 @@ interface Config {
     OPENAI: string;
     GROQ: string;
     ANTHROPIC: string;
+    AZURE: string;
   };
   API_ENDPOINTS: {
     SEARXNG: string;
     OLLAMA: string;
+    AZURE: string;
+  };
+  AZURE_OPENAI: {
+    DEPLOYMENT_NAME_GPT35_TURBO: string;
+    DEPLOYMENT_NAME_GPT35_TURBO_16K: string;
+    DEPLOYMENT_NAME_GPT35_TURBO_INSTRUCT: string;
+    DEPLOYMENT_NAME_GPT4: string;
+    DEPLOYMENT_NAME_GPT4_32K: string;
+    DEPLOYMENT_NAME_GPT4_OMNI: string;
+    DEPLOYMENT_NAME_GPT4_OMNI_MINI: string;
+    DEPLOYMENT_NAME_EMBEDDINGS_ADA: string;
+    DEPLOYMENT_NAME_EMBEDDINGS_SMALL: string;
+    DEPLOYMENT_NAME_EMBEDDINGS_LARGE: string;
   };
 }
 
@@ -48,6 +62,40 @@ export const getSearxngApiEndpoint = () =>
 
 export const getOllamaApiEndpoint = () => loadConfig().API_ENDPOINTS.OLLAMA;
 
+export const getAzureOpenaiApiKey = () => loadConfig().API_KEYS.AZURE;
+
+export const getAzureOpenaiEndpoint = () => loadConfig().API_ENDPOINTS.AZURE;
+
+export const getAzureOpenaiDeploymentNameGpt35 = () =>
+  loadConfig().AZURE_OPENAI.DEPLOYMENT_NAME_GPT35_TURBO;
+
+export const getAzureOpenaiDeploymentNameGpt35_16K = () =>
+  loadConfig().AZURE_OPENAI.DEPLOYMENT_NAME_GPT35_TURBO_16K;
+
+export const getAzureOpenaiDeploymentNameGpt35TurboInstruct = () =>
+  loadConfig().AZURE_OPENAI.DEPLOYMENT_NAME_GPT35_TURBO_INSTRUCT;
+
+export const getAzureOpenaiDeploymentNameGpt4 = () =>
+  loadConfig().AZURE_OPENAI.DEPLOYMENT_NAME_GPT4;
+
+export const getAzureOpenaiDeploymentNameGpt4_32K = () =>
+  loadConfig().AZURE_OPENAI.DEPLOYMENT_NAME_GPT4_32K;
+
+export const getAzureOpenaiDeploymentNameGpt4o = () =>
+  loadConfig().AZURE_OPENAI.DEPLOYMENT_NAME_GPT4_OMNI;
+
+export const getAzureOpenaiDeploymentNameGpt4oMini = () =>
+  loadConfig().AZURE_OPENAI.DEPLOYMENT_NAME_GPT4_OMNI_MINI;
+
+export const getAzureOpenaiDeploymentNameEmbeddingsAda = () =>
+  loadConfig().AZURE_OPENAI.DEPLOYMENT_NAME_EMBEDDINGS_ADA;
+
+export const getAzureOpenaiDeploymentNameEmbeddingsSmall = () =>
+  loadConfig().AZURE_OPENAI.DEPLOYMENT_NAME_EMBEDDINGS_SMALL;
+
+export const getAzureOpenaiDeploymentNameEmbeddingsLarge = () =>
+  loadConfig().AZURE_OPENAI.DEPLOYMENT_NAME_EMBEDDINGS_LARGE;
+
 export const updateConfig = (config: RecursivePartial<Config>) => {
   const currentConfig = loadConfig();
 

src/lib/providers/azure.ts

@@ -0,0 +1,113 @@
+import {
+  AzureChatOpenAI,
+  AzureOpenAIEmbeddings,
+} from '@langchain/azure-openai';
+import { getAzureOpenaiApiKey, getAzureOpenaiEndpoint } from '../../config';
+import {
+  getAzureOpenaiDeploymentNameGpt35,
+  getAzureOpenaiDeploymentNameGpt35_16K,
+  getAzureOpenaiDeploymentNameGpt35TurboInstruct,
+  getAzureOpenaiDeploymentNameGpt4,
+  getAzureOpenaiDeploymentNameGpt4_32K,
+  getAzureOpenaiDeploymentNameGpt4o,
+  getAzureOpenaiDeploymentNameGpt4oMini,
+  getAzureOpenaiDeploymentNameEmbeddingsAda,
+  getAzureOpenaiDeploymentNameEmbeddingsSmall,
+  getAzureOpenaiDeploymentNameEmbeddingsLarge,
+} from '../../config';
+import logger from '../../utils/logger';
+
+export const loadAzureOpenAIChatModels = async () => {
+  const azureApiKey = getAzureOpenaiApiKey();
+  const azureEndpoint = getAzureOpenaiEndpoint();
+
+  if (!azureApiKey || !azureEndpoint) {
+    logger.error('Azure OpenAI API key or endpoint is missing');
+    return {};
+  }
+
+  const chatModels = {};
+
+  try {
+    const deploymentNames = {
+      'GPT-3.5 turbo': getAzureOpenaiDeploymentNameGpt35(),
+      'GPT-3.5 turbo 16K': getAzureOpenaiDeploymentNameGpt35_16K(),
+      'GPT-3.5 turbo instruct':
+        getAzureOpenaiDeploymentNameGpt35TurboInstruct(),
+      'GPT-4': getAzureOpenaiDeploymentNameGpt4(),
+      'GPT-4 32K': getAzureOpenaiDeploymentNameGpt4_32K(),
+      'GPT-4 omni': getAzureOpenaiDeploymentNameGpt4o(),
+      'GPT-4 omni mini': getAzureOpenaiDeploymentNameGpt4oMini(),
+    };
+
+    let atLeastOneModelLoaded = false;
+
+    for (const [modelName, deploymentName] of Object.entries(deploymentNames)) {
+      try {
+        if (deploymentName) {
+          chatModels[modelName] = new AzureChatOpenAI({
+            azureOpenAIEndpoint: azureEndpoint,
+            azureOpenAIApiKey: azureApiKey,
+            azureOpenAIApiDeploymentName: deploymentName,
+            temperature: 0.7,
+          });
+          atLeastOneModelLoaded = true;
+        }
+      } catch (error) {
+        logger.warn(
+          `Deployment for ${modelName} is missing or failed to load:`,
+          error,
+        );
+        // Continue the loop even if a specific model fails
+      }
+    }
+
+    // Check if at least one model is loaded, log error if not
+    if (!atLeastOneModelLoaded) {
+      logger.error(
+        'No chat models for Azure OpenAI could be loaded. At least one model is required.',
+      );
+      return {};
+    }
+
+    return chatModels;
+  } catch (error) {
+    logger.error('Failed to load Azure OpenAI chat models', error);
+    return {};
+  }
+};
+
+export const loadAzureOpenAIEmbeddings = async () => {
+  const azureApiKey = getAzureOpenaiApiKey();
+  const azureEndpoint = getAzureOpenaiEndpoint();
+
+  if (!azureApiKey || !azureEndpoint) return {};
+
+  try {
+    const embeddings = {
+      'Text embedding Ada 002': new AzureOpenAIEmbeddings({
+        azureOpenAIEndpoint: azureEndpoint,
+        azureOpenAIApiKey: azureApiKey,
+        azureOpenAIApiDeploymentName:
+          getAzureOpenaiDeploymentNameEmbeddingsAda(),
+      }),
+      'Text embedding 3 small': new AzureOpenAIEmbeddings({
+        azureOpenAIEndpoint: azureEndpoint,
+        azureOpenAIApiKey: azureApiKey,
+        azureOpenAIApiDeploymentName:
+          getAzureOpenaiDeploymentNameEmbeddingsSmall(),
+      }),
+      'Text embedding 3 large': new AzureOpenAIEmbeddings({
+        azureOpenAIEndpoint: azureEndpoint,
+        azureOpenAIApiKey: azureApiKey,
+        azureOpenAIApiDeploymentName:
+          getAzureOpenaiDeploymentNameEmbeddingsLarge(),
+      }),
+    };
+
+    return embeddings;
+  } catch (error) {
+    logger.error('Failed to load Azure OpenAI embeddings', error);
+    return {};
+  }
+};

src/lib/providers/index.ts

@@ -1,11 +1,13 @@
 import { loadGroqChatModels } from './groq';
 import { loadOllamaChatModels, loadOllamaEmbeddingsModels } from './ollama';
 import { loadOpenAIChatModels, loadOpenAIEmbeddingsModels } from './openai';
+import { loadAzureOpenAIChatModels, loadAzureOpenAIEmbeddings } from './azure';
 import { loadAnthropicChatModels } from './anthropic';
 import { loadTransformersEmbeddingsModels } from './transformers';
 
 const chatModelProviders = {
   openai: loadOpenAIChatModels,
+  azure: loadAzureOpenAIChatModels,
   groq: loadGroqChatModels,
   ollama: loadOllamaChatModels,
   anthropic: loadAnthropicChatModels,
@@ -13,6 +15,7 @@ const chatModelProviders = {
 
 const embeddingModelProviders = {
   openai: loadOpenAIEmbeddingsModels,
+  azure: loadAzureOpenAIEmbeddings,
   local: loadTransformersEmbeddingsModels,
   ollama: loadOllamaEmbeddingsModels,
 };