🚀 Release v0.3.0: LangChain v1 Migration & Enhanced Deep Agent

README.md

@@ -2,9 +2,9 @@
 
 **LangGraph-UP Monorepo** showcases how to build production-ready LangGraph agents using the latest **LangChain & LangGraph** V1 ecosystem, organized in a clean monorepo structure with shared libraries and multiple agent applications.
 
-[![Version](https://img.shields.io/badge/version-v0.2.0-blue.svg)](https://github.com/webup/langgraph-up-monorepo/releases/tag/v0.2.0)
-[![LangChain](https://img.shields.io/badge/LangChain-v1alpha-blue.svg)](https://github.com/langchain-ai/langchain)
-[![LangGraph](https://img.shields.io/badge/LangGraph-v1alpha-blue.svg)](https://github.com/langchain-ai/langgraph)
+[![Version](https://img.shields.io/badge/version-v0.3.0-blue.svg)](https://github.com/webup/langgraph-up-monorepo/releases/tag/v0.3.0)
+[![LangChain](https://img.shields.io/badge/LangChain-v1.0-blue.svg)](https://github.com/langchain-ai/langchain)
+[![LangGraph](https://img.shields.io/badge/LangGraph-v1.0-blue.svg)](https://github.com/langchain-ai/langgraph)
 [![License](https://img.shields.io/badge/license-MIT-green.svg)](https://opensource.org/licenses/MIT)
 [![PyPI](https://img.shields.io/badge/PyPI-langgraph--up--devkits-blue.svg)](https://test.pypi.org/project/langgraph-up-devkits/)
 [![Twitter](https://img.shields.io/twitter/follow/zhanghaili0610?style=social)](https://twitter.com/zhanghaili0610)
@@ -13,7 +13,8 @@
 
 - 🌐 **Universal Model Loading** - OpenRouter, Qwen, QwQ, SiliconFlow with automatic registration
 - 🤖 **Multi-Agent Orchestration** - Supervisor & deep research patterns with specialized sub-agents
-- 🛠 **Custom Middleware** - Model switching, file masking, summarization, and state management
+- 🛠 **LangChain v1.0 Middleware** - Model switching, file masking, summarization with v1.0 pattern
+- 🔬 **Deep Research Agent** - Advanced research workflow with deepagents integration
 - 🧪 **Developer Experience** - Hot reload, comprehensive testing, strict linting, PyPI publishing
 - 🚀 **Deployment Ready** - LangGraph Cloud configurations included
 - 🌍 **Global Ready** - Region-based provider configuration (PRC/International)
@@ -154,17 +155,17 @@ math_to_research = create_handoff_tool("research_expert")
 research_to_math = create_handoff_tool("math_expert")
 ```
 
-#### 🔧 Custom Middleware (LangChain v1)
+#### 🔧 Custom Middleware (LangChain v1.0)
 
-Built-in middleware for dynamic model switching, state management, and behavior modification:
+Built-in middleware for dynamic model switching, state management, and behavior modification using the **LangChain v1.0 middleware pattern**:
 
 ```python
 from langchain.agents import create_agent
-from langgraph_up_devkits import (
+from langgraph_up_devkits.middleware import (
     ModelProviderMiddleware,
     FileSystemMaskMiddleware,
-    load_chat_model
 )
+from langgraph_up_devkits import load_chat_model
 
 # Model provider middleware for automatic switching
 model_middleware = ModelProviderMiddleware()
@@ -183,11 +184,16 @@ context = {"model": "siliconflow:Qwen/Qwen3-8B"}
 result = await agent.ainvoke(messages, context=context)
 ```
 
-**Available Middleware:**
+**Available Middleware (v1.0 Compatible):**
 - `ModelProviderMiddleware` - Dynamic model switching based on context
 - `FileSystemMaskMiddleware` - Masks virtual file systems from LLM to save tokens
 - `SummarizationMiddleware` - Automatic message summarization for long conversations
 
+**Key Changes in v1.0:**
+- Migrated to LangChain v1.0 middleware pattern with `before_model()` and `after_model()` hooks
+- Compatible with `langchain.agents.create_agent` middleware system
+- Improved state management and model switching reliability
+
 For detailed documentation on additional features like middleware, tools, and utilities, see:
 
 - **Framework Documentation**: [`libs/langgraph-up-devkits/README.md`](libs/langgraph-up-devkits/README.md)

apps/sample-deep-agent/pyproject.toml

@@ -9,7 +9,7 @@ readme = "README.md"
 license = { text = "MIT" }
 requires-python = ">=3.11,<4.0"
 dependencies = [
-    "deepagents>=0.0.11",
+    "deepagents>=0.1.1",
     "langgraph-up-devkits",
 ]
 

apps/sample-deep-agent/src/sample_deep_agent/context.py

@@ -10,7 +10,7 @@ class DeepAgentContext(BaseModel):
     """Context configuration for deep agent runtime settings."""
 
     # Model configuration
-    model_name: str = Field(default="siliconflow:zai-org/GLM-4.5", description="Default model name")
+    model_name: str = Field(default="siliconflow:deepseek-ai/DeepSeek-V3.2-Exp", description="Default model name")
 
     # Graph configuration
     recursion_limit: int = Field(default=1000, description="Recursion limit for agent execution")

apps/sample-deep-agent/src/sample_deep_agent/graph.py

@@ -2,7 +2,7 @@
 
 from typing import Any
 
-from deepagents import async_create_deep_agent  # type: ignore[import-untyped]
+from deepagents import create_deep_agent  # type: ignore[import-untyped]
 from langchain_core.runnables import RunnableConfig
 from langgraph_up_devkits import load_chat_model
 from langgraph_up_devkits.tools import deep_web_search, think_tool
@@ -35,12 +35,12 @@ def make_graph(config: RunnableConfig | None = None) -> Any:
     # Load model based on context configuration
     model = load_chat_model(context.model_name)
 
-    # Create deep agent with research capabilities (remove research_sub_agent from subagents list)
-    agent = async_create_deep_agent(
-        tools=[deep_web_search, think_tool],
-        instructions=get_research_instructions(),
-        subagents=RESEARCH_AGENTS,  # Research agent in subagents list
+    # Create deep agent with research capabilities
+    agent = create_deep_agent(
         model=model,
+        tools=[deep_web_search, think_tool],
+        system_prompt=get_research_instructions(),
+        subagents=RESEARCH_AGENTS,
         context_schema=DeepAgentContext,
     ).with_config({"recursion_limit": context.recursion_limit})
 

apps/sample-deep-agent/src/sample_deep_agent/subagents.py

@@ -17,7 +17,7 @@
         "into the necessary components, and then call multiple research agents in parallel, "
         "one for each sub question."
     ),
-    "prompt": SUB_RESEARCH_PROMPT,
+    "system_prompt": SUB_RESEARCH_PROMPT,
     "middleware": [filesystem_mask],
 }
 
@@ -27,7 +27,7 @@
         "Used to critique the final report. Give this agent some information about "
         "how you want it to critique the report."
     ),
-    "prompt": SUB_CRITIQUE_PROMPT,
+    "system_prompt": SUB_CRITIQUE_PROMPT,
     "middleware": [filesystem_mask],
 }
 

apps/sample-deep-agent/tests/conftest.py

@@ -5,9 +5,13 @@
 import pytest
 
 
-@pytest.fixture(autouse=True)
+@pytest.fixture
 def mock_openai_key(monkeypatch):
-    """Mock API keys to prevent API key errors in unit tests."""
+    """Mock API keys to prevent API key errors in unit tests.
+
+    Note: This fixture is NOT autouse - it must be explicitly requested by unit tests.
+    Integration tests need real API keys from the environment.
+    """
     monkeypatch.setenv("OPENAI_API_KEY", "sk-test-fake-key-for-unit-tests")
     monkeypatch.setenv("OPENROUTER_API_KEY", "sk-test-fake-key-for-unit-tests")
     monkeypatch.setenv("SILICONFLOW_API_KEY", "sk-test-fake-key-for-unit-tests")

apps/sample-deep-agent/tests/unit/test_graph.py

@@ -10,7 +10,7 @@
 class TestGraphCreation:
     """Test graph creation and configuration."""
 
-    @patch('sample_deep_agent.graph.async_create_deep_agent')
+    @patch('sample_deep_agent.graph.create_deep_agent')
     @patch('sample_deep_agent.graph.load_chat_model')
     def test_make_graph_with_default_config(self, mock_load_model, mock_create_deep_agent):
         """Test graph creation with default configuration."""
@@ -25,7 +25,7 @@ def test_make_graph_with_default_config(self, mock_load_model, mock_create_deep_
 
         result = make_graph()
 
-        mock_load_model.assert_called_once_with("siliconflow:zai-org/GLM-4.5")
+        mock_load_model.assert_called_once_with("siliconflow:deepseek-ai/DeepSeek-V3.2-Exp")
         mock_create_deep_agent.assert_called_once()
 
         call_args = mock_create_deep_agent.call_args
@@ -40,7 +40,7 @@ def test_make_graph_with_default_config(self, mock_load_model, mock_create_deep_
 
         assert result == mock_agent_with_config
 
-    @patch('sample_deep_agent.graph.async_create_deep_agent')
+    @patch('sample_deep_agent.graph.create_deep_agent')
     @patch('sample_deep_agent.graph.load_chat_model')
     def test_make_graph_with_custom_config(self, mock_load_model, mock_create_deep_agent):
         """Test graph creation with custom configuration."""
@@ -67,14 +67,14 @@ def test_make_graph_with_custom_config(self, mock_load_model, mock_create_deep_a
 
         call_args = mock_create_deep_agent.call_args
 
-        # Verify instructions are present (runtime context will be used during actual execution)
-        instructions = call_args[1]['instructions']
-        assert "expert research coordinator" in instructions.lower()
-        assert "TODOs" in instructions
+        # Verify system_prompt is present (runtime context will be used during actual execution)
+        system_prompt = call_args[1]['system_prompt']
+        assert "expert research coordinator" in system_prompt.lower()
+        assert "TODOs" in system_prompt
 
         assert result == mock_agent_with_config
 
-    @patch('sample_deep_agent.graph.async_create_deep_agent')
+    @patch('sample_deep_agent.graph.create_deep_agent')
     @patch('sample_deep_agent.graph.load_chat_model')
     def test_make_graph_with_none_config(self, mock_load_model, mock_create_deep_agent):
         """Test graph creation with None config."""
@@ -90,12 +90,12 @@ def test_make_graph_with_none_config(self, mock_load_model, mock_create_deep_age
         result = make_graph(None)
 
         # Should use default configuration
-        mock_load_model.assert_called_once_with("siliconflow:zai-org/GLM-4.5")
+        mock_load_model.assert_called_once_with("siliconflow:deepseek-ai/DeepSeek-V3.2-Exp")
         mock_create_deep_agent.assert_called_once()
 
         assert result == mock_agent_with_config
 
-    @patch('sample_deep_agent.graph.async_create_deep_agent')
+    @patch('sample_deep_agent.graph.create_deep_agent')
     @patch('sample_deep_agent.graph.load_chat_model')
     def test_tools_are_included(self, mock_load_model, mock_create_deep_agent):
         """Test that required tools are included."""
@@ -156,36 +156,36 @@ class TestSubAgentConfiguration:
     def test_subagent_configuration_valid(self):
         """Test that sub-agent configuration is valid for deepagents."""
         # Verify all required fields are present
-        required_fields = ["name", "description", "prompt"]
+        required_fields = ["name", "description", "system_prompt"]
         for field in required_fields:
             assert field in research_sub_agent
 
         # Research agent no longer has explicit tools - they're passed from main agent
-        # Check that the prompt contains TODO constraints instead
+        # Check that the system_prompt contains TODO constraints instead
 
         # Verify name and description
         assert research_sub_agent["name"] == "research-agent"
         assert "research" in research_sub_agent["description"].lower()
 
-        # Verify prompt contains expected elements
-        prompt = research_sub_agent["prompt"]
-        assert "researcher" in prompt.lower()
-        assert "TODO CONSTRAINTS" in prompt
-        assert "GLOBAL TODO LIMIT" in prompt
-        assert str(3) in prompt  # MAX_TODOS value
+        # Verify system_prompt contains expected elements
+        system_prompt = research_sub_agent["system_prompt"]
+        assert "researcher" in system_prompt.lower()
+        assert "TODO CONSTRAINTS" in system_prompt
+        assert "GLOBAL TODO LIMIT" in system_prompt
+        assert str(3) in system_prompt  # MAX_TODOS value
 
     def test_critique_agent_configuration(self):
         """Test critique agent configuration."""
         assert critique_sub_agent["name"] == "critique-agent"
         assert "critique" in critique_sub_agent["description"].lower()
-        assert "prompt" in critique_sub_agent
+        assert "system_prompt" in critique_sub_agent
 
         # Critique agent no longer has explicit tools - they're passed from main agent
-        # Check that the prompt contains TODO constraints
-        prompt = critique_sub_agent["prompt"]
-        assert "TODO CONSTRAINTS" in prompt
-        assert "GLOBAL TODO LIMIT" in prompt
-        assert str(3) in prompt  # MAX_TODOS value
+        # Check that the system_prompt contains TODO constraints
+        system_prompt = critique_sub_agent["system_prompt"]
+        assert "TODO CONSTRAINTS" in system_prompt
+        assert "GLOBAL TODO LIMIT" in system_prompt
+        assert str(3) in system_prompt  # MAX_TODOS value
 
 
 class TestAppExport:
@@ -228,9 +228,9 @@ def test_tool_integration_in_graph(self):
         assert hasattr(deep_web_search, "invoke") or callable(deep_web_search)
         assert hasattr(think_tool, "invoke") or callable(think_tool)
 
-    @patch('sample_deep_agent.graph.async_create_deep_agent')
+    @patch('sample_deep_agent.graph.create_deep_agent')
     def test_tools_passed_to_deep_agent(self, mock_create_deep_agent):
-        """Test that tools are properly passed to async_create_deep_agent."""
+        """Test that tools are properly passed to create_deep_agent."""
         from sample_deep_agent.graph import make_graph
 
         mock_agent = Mock()