diff --git a/README.md b/README.md
index 184db529..48a18a38 100644
--- a/README.md
+++ b/README.md
@@ -68,7 +68,7 @@ Easy LLM CLI supports connecting to any OpenAI-compatible LLM API. You can confi
 # Enable custom LLM support
 export USE_CUSTOM_LLM=true 
 
-export CUSTOM_LLM_PROVIDER="openai"  # LLM provider
+export CUSTOM_LLM_PROVIDER="openai"  # LLM provider, if you use azure openai, set to "azure"
 export CUSTOM_LLM_API_KEY="your-api-key"     # Your LLM provider API key
 export CUSTOM_LLM_ENDPOINT="https://api.your-llm-provider.com/v1"  # API endpoint
 export CUSTOM_LLM_MODEL_NAME="your-model-name"  # Model name
diff --git a/README.zh-CN.md b/README.zh-CN.md
index 7d974af9..6964035f 100644
--- a/README.zh-CN.md
+++ b/README.zh-CN.md
@@ -66,6 +66,8 @@ Easy LLM 命令行工具支持连接到任何兼容 OpenAI 的 LLM API。你可
 ```bash
 # 启用自定义 LLM 支持
 export USE_CUSTOM_LLM=true
+
+export CUSTOM_LLM_PROVIDER="openai"  # LLM provider, 如果你使用Azure OpneAI, set to "azure"
 export CUSTOM_LLM_API_KEY="your-api-key"     # 你的 LLM 提供商 API 密钥
 export CUSTOM_LLM_ENDPOINT="https://api.your-llm-provider.com/v1"  # API 端点
 export CUSTOM_LLM_MODEL_NAME="your-model-name"  # 模型名称
diff --git a/packages/core/src/custom_llm/converter.ts b/packages/core/src/custom_llm/converter.ts
index a26ed104..fa33c2ce 100644
--- a/packages/core/src/custom_llm/converter.ts
+++ b/packages/core/src/custom_llm/converter.ts
@@ -341,7 +341,7 @@ export class ModelConverter {
       }
     }
 
-    if (choice.finish_reason === 'tool_calls' && toolCallMap.size > 0) {
+    if (choice?.finish_reason === 'tool_calls' && toolCallMap.size > 0) {
       const response = this.toGeminiStreamToolCallsResponse(toolCallMap);
       toolCallMap.clear();
       return {
diff --git a/packages/core/src/custom_llm/index.ts b/packages/core/src/custom_llm/index.ts
index f36a5186..d214a8d7 100644
--- a/packages/core/src/custom_llm/index.ts
+++ b/packages/core/src/custom_llm/index.ts
@@ -12,7 +12,7 @@ import {
   EmbedContentResponse,
   EmbedContentParameters,
 } from '@google/genai';
-import OpenAI from 'openai';
+import { OpenAI, AzureOpenAI } from 'openai';
 import { ContentGenerator } from '../core/contentGenerator.js';
 import { CustomLLMContentGeneratorConfig, ToolCallMap } from './types.js';
 import { extractToolFunctions } from './util.js';
@@ -23,6 +23,7 @@ export class CustomLLMContentGenerator implements ContentGenerator {
   private apiKey: string = process.env.CUSTOM_LLM_API_KEY || '';
   private baseURL: string = process.env.CUSTOM_LLM_ENDPOINT || '';
   private modelName: string = process.env.CUSTOM_LLM_MODEL_NAME || '';
+  private apiVersion: string = process.env.OPENAI_API_VERSION || '2024-12-01-preview';
   private temperature: number = Number(process.env.CUSTOM_LLM_TEMPERATURE || 0);
   private maxTokens: number = Number(process.env.CUSTOM_LLM_MAX_TOKENS || 8192);
   private topP: number = Number(process.env.CUSTOM_LLM_TOP_P || 1);
@@ -34,11 +35,20 @@ export class CustomLLMContentGenerator implements ContentGenerator {
   };
 
   constructor() {
-    this.model = new OpenAI({
-      apiKey: this.apiKey,
-      baseURL: this.baseURL,
-    });
-  }
+      const provider = process.env.CUSTOM_LLM_PROVIDER || 'openai';
+      if (provider === 'azure') {
+        this.model = new AzureOpenAI({
+          apiKey: this.apiKey,
+          endpoint: this.baseURL,
+          apiVersion: this.apiVersion
+        });
+      } else {
+        this.model = new OpenAI({
+          apiKey: this.apiKey,
+          baseURL: this.baseURL
+        });
+      }
+    }
 
   /**
    * Asynchronously generates content responses in a streaming fashion.