Fix: Azure OpenAI model key collision (#5092)

* fix: Migrate AzureOpenAI model key from OPEN_MODEL_PREF to prevent the naming collision. No effort necessary from current users.

* test: add backwards compat tests for AzureOpenAI model key migration

* patch missing env example file

* linting

---------

Co-authored-by: Timothy Carambat <rambat1010@gmail.com>

docker/.env.example

@@ -21,7 +21,7 @@ GID='1000'
 # LLM_PROVIDER='azure'
 # AZURE_OPENAI_ENDPOINT=
 # AZURE_OPENAI_KEY=
-# OPEN_MODEL_PREF='my-gpt35-deployment' # This is the "deployment" on Azure you want to use. Not the base model.
+# AZURE_OPENAI_MODEL_PREF='my-gpt35-deployment' # This is the "deployment" on Azure you want to use. Not the base model.
 # EMBEDDING_MODEL_PREF='embedder-model' # This is the "deployment" on Azure you want to use for embeddings. Not the base model. Valid base model is text-embedding-ada-002
 
 # LLM_PROVIDER='anthropic'

server/.env.example

@@ -18,7 +18,7 @@ SIG_SALT='salt' # Please generate random string at least 32 chars long.
 # LLM_PROVIDER='azure'
 # AZURE_OPENAI_ENDPOINT=
 # AZURE_OPENAI_KEY=
-# OPEN_MODEL_PREF='my-gpt35-deployment' # This is the "deployment" on Azure you want to use. Not the base model.
+# AZURE_OPENAI_MODEL_PREF='my-gpt35-deployment' # This is the "deployment" on Azure you want to use. Not the base model.
 # EMBEDDING_MODEL_PREF='embedder-model' # This is the "deployment" on Azure you want to use for embeddings. Not the base model. Valid base model is text-embedding-ada-002
 
 # LLM_PROVIDER='anthropic'

server/__tests__/utils/helpers/azureOpenAiModelPref.test.js

@@ -0,0 +1,55 @@
+/* eslint-env jest */
+
+/**
+ * Tests for the AzureOpenAI model key migration from OPEN_MODEL_PREF
+ * to AZURE_OPENAI_MODEL_PREF, ensuring backwards compatibility for
+ * existing users who have OPEN_MODEL_PREF set.
+ *
+ * Related issue: https://github.com/Mintplex-Labs/anything-llm/issues/3839
+ */
+
+describe("AzureOpenAI model key backwards compatibility", () => {
+  const ORIGINAL_ENV = process.env;
+
+  beforeEach(() => {
+    jest.resetModules();
+    process.env = { ...ORIGINAL_ENV };
+    delete process.env.AZURE_OPENAI_MODEL_PREF;
+    delete process.env.OPEN_MODEL_PREF;
+  });
+
+  afterAll(() => {
+    process.env = ORIGINAL_ENV;
+  });
+
+  describe("getBaseLLMProviderModel - helpers/index.js", () => {
+    test("returns AZURE_OPENAI_MODEL_PREF when set", () => {
+      process.env.AZURE_OPENAI_MODEL_PREF = "my-azure-deployment";
+      process.env.OPEN_MODEL_PREF = "gpt-4o";
+      const { getBaseLLMProviderModel } = require("../../../utils/helpers/index");
+      expect(getBaseLLMProviderModel({ provider: "azure" })).toBe("my-azure-deployment");
+    });
+
+    test("falls back to OPEN_MODEL_PREF when AZURE_OPENAI_MODEL_PREF is not set (backwards compat)", () => {
+      process.env.OPEN_MODEL_PREF = "my-old-azure-deployment";
+      const { getBaseLLMProviderModel } = require("../../../utils/helpers/index");
+      expect(getBaseLLMProviderModel({ provider: "azure" })).toBe("my-old-azure-deployment");
+    });
+
+    test("openai provider still uses OPEN_MODEL_PREF exclusively", () => {
+      process.env.OPEN_MODEL_PREF = "gpt-4o";
+      process.env.AZURE_OPENAI_MODEL_PREF = "my-azure-deployment";
+      const { getBaseLLMProviderModel } = require("../../../utils/helpers/index");
+      expect(getBaseLLMProviderModel({ provider: "openai" })).toBe("gpt-4o");
+    });
+
+    test("azure and openai return different values when both keys are set", () => {
+      process.env.OPEN_MODEL_PREF = "gpt-4o";
+      process.env.AZURE_OPENAI_MODEL_PREF = "my-azure-deployment";
+      const { getBaseLLMProviderModel } = require("../../../utils/helpers/index");
+      expect(getBaseLLMProviderModel({ provider: "azure" })).not.toBe(
+        getBaseLLMProviderModel({ provider: "openai" })
+      );
+    });
+  });
+});

server/endpoints/utils.js

@@ -97,7 +97,8 @@ function getModelTag() {
       model = process.env.TOGETHER_AI_MODEL_PREF;
       break;
     case "azure":
-      model = process.env.OPEN_MODEL_PREF;
+      model =
+        process.env.AZURE_OPENAI_MODEL_PREF || process.env.OPEN_MODEL_PREF;
       break;
     case "koboldcpp":
       model = process.env.KOBOLD_CPP_MODEL_PREF;

server/models/systemSettings.js

@@ -508,7 +508,8 @@ const SystemSettings = {
       // Azure + OpenAI Keys
       AzureOpenAiEndpoint: process.env.AZURE_OPENAI_ENDPOINT,
       AzureOpenAiKey: !!process.env.AZURE_OPENAI_KEY,
-      AzureOpenAiModelPref: process.env.OPEN_MODEL_PREF,
+      AzureOpenAiModelPref:
+        process.env.AZURE_OPENAI_MODEL_PREF || process.env.OPEN_MODEL_PREF,
       AzureOpenAiEmbeddingModelPref: process.env.EMBEDDING_MODEL_PREF,
       AzureOpenAiTokenLimit: process.env.AZURE_OPENAI_TOKEN_LIMIT || 4096,
       AzureOpenAiModelType: process.env.AZURE_OPENAI_MODEL_TYPE || "default",

server/utils/AiProviders/azureOpenAi/index.js

@@ -20,7 +20,10 @@ class AzureOpenAiLLM {
       apiKey: process.env.AZURE_OPENAI_KEY,
       baseURL: AzureOpenAiLLM.formatBaseUrl(process.env.AZURE_OPENAI_ENDPOINT),
     });
-    this.model = modelPreference ?? process.env.OPEN_MODEL_PREF;
+    this.model =
+      modelPreference ||
+      process.env.AZURE_OPENAI_MODEL_PREF ||
+      process.env.OPEN_MODEL_PREF;
     /* 
       Note: Azure OpenAI deployments do not expose model metadata that would allow us to
       programmatically detect whether the deployment uses a reasoning model (o1, o1-mini, o3-mini, etc.).
@@ -150,7 +153,7 @@ class AzureOpenAiLLM {
   async getChatCompletion(messages = [], { temperature = 0.7 }) {
     if (!this.model)
       throw new Error(
-        "No OPEN_MODEL_PREF ENV defined. This must the name of a deployment on your Azure account for an LLM chat model like GPT-3.5."
+        "No AZURE_OPENAI_MODEL_PREF ENV defined. This must the name of a deployment on your Azure account for an LLM chat model like GPT-3.5."
       );
 
     const result = await LLMPerformanceMonitor.measureAsyncFunction(
@@ -185,7 +188,7 @@ class AzureOpenAiLLM {
   async streamGetChatCompletion(messages = [], { temperature = 0.7 }) {
     if (!this.model)
       throw new Error(
-        "No OPEN_MODEL_PREF ENV defined. This must the name of a deployment on your Azure account for an LLM chat model like GPT-3.5."
+        "No AZURE_OPENAI_MODEL_PREF ENV defined. This must the name of a deployment on your Azure account for an LLM chat model like GPT-3.5."
       );
 
     const measuredStreamRequest = await LLMPerformanceMonitor.measureStream({

server/utils/agents/aibitat/providers/azure.js

@@ -17,7 +17,10 @@ class AzureOpenAiProvider extends Provider {
       baseURL: AzureOpenAiLLM.formatBaseUrl(process.env.AZURE_OPENAI_ENDPOINT),
     });
     super(client);
-    this.model = config.model ?? process.env.OPEN_MODEL_PREF;
+    this.model =
+      config.model ||
+      process.env.AZURE_OPENAI_MODEL_PREF ||
+      process.env.OPEN_MODEL_PREF;
     this.verbose = true;
   }
 

server/utils/agents/index.js

@@ -268,7 +268,9 @@ class AgentHandler {
           "mistralai/Mixtral-8x7B-Instruct-v0.1"
         );
       case "azure":
-        return process.env.OPEN_MODEL_PREF;
+        return (
+          process.env.AZURE_OPENAI_MODEL_PREF || process.env.OPEN_MODEL_PREF
+        );
       case "koboldcpp":
         return process.env.KOBOLD_CPP_MODEL_PREF ?? null;
       case "localai":

server/utils/helpers/index.js

@@ -440,7 +440,7 @@ function getBaseLLMProviderModel({ provider = null } = {}) {
     case "openai":
       return process.env.OPEN_MODEL_PREF;
     case "azure":
-      return process.env.OPEN_MODEL_PREF;
+      return process.env.AZURE_OPENAI_MODEL_PREF || process.env.OPEN_MODEL_PREF;
     case "anthropic":
       return process.env.ANTHROPIC_MODEL_PREF;
     case "gemini":

server/utils/helpers/updateENV.js

@@ -32,7 +32,7 @@ const KEY_MAPPING = {
     checks: [isNotEmpty],
   },
   AzureOpenAiModelPref: {
-    envKey: "OPEN_MODEL_PREF",
+    envKey: "AZURE_OPENAI_MODEL_PREF",
     checks: [isNotEmpty],
   },
   AzureOpenAiEmbeddingModelPref: {