feat: add azure ai llm provider

.env.example

@@ -8,6 +8,11 @@ GOOGLE_API_KEY="..."
 XAI_API_KEY='...'
 OPEN_ROUTER_API_KEY='..'
 
+# Azure AI Foundry
+# With API key:     AZURE_BASE_URL="https://<resource>.services.ai.azure.com/openai/v1"
+# With Entra ID:    AZURE_BASE_URL="https://<resource>.services.ai.azure.com/api/projects/<project>"
+# AZURE_API_KEY="..." # Optional - omit to use DefaultAzureCredential
+
 # EVENTS_OUTPUT_PATH="./events.txt" # Optional - path to output events file
 # RESULTS_OUTPUT_PATH="./results.txt" # Optional - path to output results file
 

llm-config.override.template.jsonc

@@ -1,6 +1,6 @@
 // Override the LLM config here for the agents.
 
-// Supported providers: openai, google, openrouter, xai, vertexai
+// Supported providers: openai, google, openrouter, xai, vertexai, azure
 // You can use any model as long as it is supported by the provider.
 
 // Do not modify this file directly, instead copy it to llm.override.jsonc and modify it there.

minitap/mobile_use/agents/outputter/test_outputter.py

@@ -10,6 +10,8 @@
 sys.modules["langchain_google_genai"] = Mock()
 sys.modules["langchain_openai"] = Mock()
 sys.modules["langchain_cerebras"] = Mock()
+sys.modules["langchain_azure_ai"] = Mock()
+sys.modules["langchain_azure_ai.chat_models"] = Mock()
 
 from minitap.mobile_use.agents.outputter.outputter import outputter  # noqa: E402
 from minitap.mobile_use.config import LLM, OutputConfig  # noqa: E402

minitap/mobile_use/config.py

@@ -23,9 +23,11 @@ class Settings(BaseSettings):
     GOOGLE_API_KEY: SecretStr | None = None
     XAI_API_KEY: SecretStr | None = None
     OPEN_ROUTER_API_KEY: SecretStr | None = None
+    AZURE_API_KEY: SecretStr | None = None
     MINITAP_API_KEY: SecretStr | None = None
 
     OPENAI_BASE_URL: str | None = None
+    AZURE_BASE_URL: str | None = None
     MINITAP_BASE_URL: str = "https://platform.minitap.ai"
 
     ADB_HOST: str | None = None
@@ -94,7 +96,7 @@ def record_events(output_path: Path | None, events: list[str] | BaseModel | Any)
 
 ### LLM Configuration
 
-LLMProvider = Literal["openai", "google", "openrouter", "xai", "vertexai", "minitap"]
+LLMProvider = Literal["openai", "google", "openrouter", "xai", "vertexai", "minitap", "azure"]
 LLMUtilsNode = Literal["outputter", "hopper", "video_analyzer"]
 LLMUtilsNodeWithFallback = LLMUtilsNode
 AgentNode = Literal[
@@ -128,6 +130,9 @@ class LLM(BaseModel):
 
     def validate_provider(self, name: str):
         match self.provider:
+            case "azure":
+                if not settings.AZURE_BASE_URL:
+                    raise Exception(f"{name} requires AZURE_BASE_URL in .env")
             case "openai":
                 if not settings.OPENAI_API_KEY:
                     raise Exception(f"{name} requires OPENAI_API_KEY in .env")
@@ -146,6 +151,7 @@ def validate_provider(self, name: str):
                 if not settings.MINITAP_API_KEY:
                     raise Exception(f"{name} requires MINITAP_API_KEY in .env")
 
+
     def __str__(self):
         return f"{self.provider}/{self.model}"
 

minitap/mobile_use/services/llm.py

@@ -4,6 +4,7 @@
 from typing import Any, Literal, TypeVar, overload
 
 from langchain_core.language_models.chat_models import BaseChatModel
+from langchain_azure_ai.chat_models import AzureAIOpenAIApiChatModel
 from langchain_google_genai import ChatGoogleGenerativeAI
 from langchain_google_vertexai import ChatVertexAI
 from langchain_openai import ChatOpenAI
@@ -169,6 +170,31 @@ def get_grok_llm(model_name: str, temperature: float = 1) -> ChatOpenAI:
     )
     return client
 
+def get_azure_llm(model_name: str, temperature: float = 1) -> AzureAIOpenAIApiChatModel:
+    assert settings.AZURE_BASE_URL is not None
+
+    api_key: str | None = None
+    if settings.AZURE_API_KEY:
+        api_key = settings.AZURE_API_KEY.get_secret_value()
+
+    if api_key:
+        client = AzureAIOpenAIApiChatModel(
+            model=model_name,
+            endpoint=settings.AZURE_BASE_URL,
+            credential=api_key,
+            temperature=temperature,
+        )
+    else:
+        from azure.identity import DefaultAzureCredential
+
+        client = AzureAIOpenAIApiChatModel(
+            model=model_name,
+            project_endpoint=settings.AZURE_BASE_URL,
+            credential=DefaultAzureCredential(),
+            temperature=temperature,
+        )
+    return client
+
 
 @overload
 def get_llm(
@@ -228,6 +254,8 @@ def get_llm(
         return get_openrouter_llm(llm.model, temperature)
     elif llm.provider == "xai":
         return get_grok_llm(llm.model, temperature)
+    elif llm.provider == "azure":
+        return get_azure_llm(llm.model, temperature)
     elif llm.provider == "minitap":
         remote_tracing = False
         if ctx.execution_setup:

pyproject.toml

@@ -42,6 +42,7 @@ dependencies = [
     "uuid-utils>=0.12.0",
     "limrun-api>=0.1.0",
     "websockets>=12.0",
+    "langchain-azure-ai>=1.1.1",
 ]
 
 [project.optional-dependencies]