LangChain DeepAgents x AI Agent A2Z Deploy LangChain Examples Live

Website | GitHub | AI Agent Marketplace | AI Agent A2Z

This guide shows how to convert a LangChain DeepAgents into a production-ready live service using BaseLiveRuntime from ai-agent-marketplace

The runtime wraps your agent and exposes a FastAPI streaming /chat endpoint automatically.

QuickStart

from ai_agent_marketplace.runtime.base import *

async def deepagents_stream_generator(agent: Any, user_query: str, **kwargs) -> AsyncGenerator[str, None]:
    """
    Universal async adapter
    """
    initial_content = "Task Started..."
    initial_chunk = json.dumps(assembly_message())
    ### 
    yield initial_chunk + STREAMING_SEPARATOR_DEFAULT

runtime = BaseLiveRuntime(
    agent=agent,
    stream_handler=deepagents_stream_generator
)

## Returned a FastAPI based app with /chat endpoint
app = runtime.app

Run the Server

Set the Environment Variables

# Set API keys
export GOOGLE_API_KEY="..."      # For image generation
export TAVILY_API_KEY="..."      # For web search (optional)

Entry Point Command shell

uvicorn langchain_deepagents.deep_research.research_agent_server:app

🚀 Architecture Summary

LangChain Agent
        ↓
Streaming Adapter (Async Generator)
        ↓
BaseLiveRuntime
        ↓
FastAPI App (/chat)
        ↓
Streaming JSON Response

2️⃣ Create a Streaming Adapter

Define an async generator that adapts your LangChain agent output into streaming chunks.

The async generator takes in two parameters: agent an customized agent object, user_query that are parsed from the messages object from the “\chat” endpoints. In the async generator, the agent calls agent.invoke({"messages": messages}) methods.

from ai_agent_marketplace.runtime.base import *
from typing import Any, AsyncGenerator
import json
import uuid
import asyncio

async def deepagents_stream_generator(
    agent: Any,
    user_query: str,
    **kwargs
) -> AsyncGenerator[str, None]:
    """
    Universal async adapter for LangChain agent
    """

    # Send initial streaming message
    initial_content = "Task Started and Research Take a Few Minutes"
    initial_chunk = json.dumps(
        assembly_message(
            type=MESSAGE_TYPE_ASSISTANT,
            format=OUTPUT_FORMAT_TEXT,
            content=initial_content,
            content_type=CONTENT_TYPE_MARKDOWN,
            section=SECTION_ANSWER,
            message_id=str(uuid.uuid4()),
            template=TEMPLATE_STREAMING_CONTENT_TYPE,
        )
    )

    yield initial_chunk + STREAMING_SEPARATOR_DEFAULT
    await asyncio.sleep(0)

    try:
        # Call LangChain agent
        messages = [{"role": "user", "content": user_query}]
        result = agent.invoke({"messages": messages})

        output_messages = result["messages"] if "messages" in result else []

        for message in output_messages:
            message_id, content, role = extract_message_content_langchain(message)

            output_chunk = json.dumps(
                assembly_message(
                    type=MESSAGE_TYPE_ASSISTANT,
                    format=OUTPUT_FORMAT_TEXT,
                    content=content,
                    content_type=CONTENT_TYPE_MARKDOWN,
                    section=SECTION_ANSWER,
                    message_id=message_id,
                    template=TEMPLATE_STREAMING_CONTENT_TYPE,
                )
            )

            yield output_chunk + STREAMING_SEPARATOR_DEFAULT

    except Exception:
        yield json.dumps({}) + STREAMING_SEPARATOR_DEFAULT

---

3️⃣ Wrap the Agent with Runtime

runtime = BaseLiveRuntime(
    agent=agent,
    stream_handler=deepagents_stream_generator
)

# FastAPI app with /chat endpoint
app = runtime.app

That’s it.

You now have a production-ready FastAPI service.

---

Deploy the Server

Step 1. Choose Github Tab Step 2. Public url: https://github.com/aiagenta2z/agent-mcp-deployment-templates Step 3. Entry Point Command shell

uvicorn langchain_deepagents.deep_research.research_agent_server:app

Step 4.

Set the Environment Variables

# Set API keys
export GOOGLE_API_KEY="..."      # For image generation
export TAVILY_API_KEY="..."      # For web search (optional)

Step 5. Click Deploy and You will get the URL

Get the Product /chat POST URL :

https://langchain-ai.aiagenta2z.com/deep_research/chat

Server will run at:

http://localhost:8000

---

5️⃣ Test with curl

Case 1: Simple Math

curl -X POST "http://localhost:8000/chat" \
  -H "Content-Type: application/json" \
  -d '{"messages":[{"role":"user","content":"Calculate 1+1 result"}]}'

Sample Streaming Output

{"type":"assistant","format":"text","content":"Task Started...","section":"answer","message_id":"670d3458-a539-406f-a786-1afc0f0fc201","content_type":"text/markdown","template":"streaming_content_type"}
{"type":"assistant","format":"text","content":"Calculate 1+1 result","section":"answer","message_id":"701be311-37e3-4ee1-9519-6d8e65b47f59","content_type":"text/markdown","template":"streaming_content_type"}
{"type":"assistant","format":"text","content":"1 + 1 = 2","section":"answer","message_id":"lc_run--019c55fe-4ed2-7da3-9e05-0a8758aa10cc-0","content_type":"text/markdown","template":"streaming_content_type"}

---

Case 2: Research Task

curl -X POST "http://localhost:8000/chat" \
  -H "Content-Type: application/json" \
  -d '{"messages":[{"role":"user","content":"research context engineering approaches used to build AI agents"}]}'

Sample Streaming Output (Truncated)

{"type":"assistant","content":"Task Started..."}
{"type":"assistant","content":"Updated todo list ..."}
{"type":"assistant","content":"Updated file /research_request.md"}
{"type":"assistant","content":"Here is a comprehensive report on context engineering approaches..."}

The response is streamed incrementally as the agent reasons, calls tools, and produces final output.

---

🌐 Deploy Example

You can also deploy publicly:

curl -X POST "https://deepagents.aiagenta2z.com/deep_research/chat" \
  -H "Content-Type: application/json" \
  -d '{"messages":[{"role":"user","content":"Calculate 1+1 result"}]}'

---

✅ What You Get

• Automatic /chat endpoint

• Streaming JSON output

• LangChain message compatibility

• Standardized message schema

• Ready for UI integration

• Production-ready FastAPI server

🚀 Deep Research

🚀 Quickstart

Prerequisites: Install uv package manager:

curl -LsSf https://astral.sh/uv/install.sh | sh

Ensure you are in the deep_research directory:

cd examples/deep_research

Install packages:

uv sync

Set your API keys in your environment:

export ANTHROPIC_API_KEY=your_anthropic_api_key_here  # Required for Claude model
export GOOGLE_API_KEY=your_google_api_key_here        # Required for Gemini model ([get one here](https://ai.google.dev/gemini-api/docs))
export TAVILY_API_KEY=your_tavily_api_key_here        # Required for web search ([get one here](https://www.tavily.com/)) with a generous free tier
export LANGSMITH_API_KEY=your_langsmith_api_key_here  # [LangSmith API key](https://smith.langchain.com/settings) (free to sign up)

Usage Options

You can run this example in two ways:

Option 1: Jupyter Notebook

Run the interactive notebook to step through the research agent:

uv run jupyter notebook research_agent.ipynb

Option 2: LangGraph Server

Run a local LangGraph server with a web interface:

langgraph dev

LangGraph server will open a new browser window with the Studio interface, which you can submit your search query to:

You can also connect the LangGraph server to a UI specifically designed for deepagents:

git clone https://github.com/langchain-ai/deep-agents-ui.git
cd deep-agents-ui
yarn install
yarn dev

Then follow the instructions in the deep-agents-ui README to connect the UI to the running LangGraph server.

This provides a user-friendly chat interface and visualization of files in state.

📚 Resources

• Deep Research Course - Full course on deep research with LangGraph

Custom Model

By default, deepagents uses "claude-sonnet-4-5-20250929". You can customize this by passing any LangChain model object. See the Deep Agents package README for more details.

from langchain.chat_models import init_chat_model
from deepagents import create_deep_agent

# Using Claude
model = init_chat_model(model="anthropic:claude-sonnet-4-5-20250929", temperature=0.0)

# Using Gemini
from langchain_google_genai import ChatGoogleGenerativeAI
model = ChatGoogleGenerativeAI(model="gemini-3-pro-preview")

agent = create_deep_agent(
    model=model,
)

Custom Instructions

The deep research agent uses custom instructions defined in research_agent/prompts.py that complement (rather than duplicate) the default middleware instructions. You can modify these in any way you want.

Instruction Set	Purpose
RESEARCH_WORKFLOW_INSTRUCTIONS	Defines the 5-step research workflow: save request → plan with TODOs → delegate to sub-agents → synthesize → respond. Includes research-specific planning guidelines like batching similar tasks and scaling rules for different query types.
SUBAGENT_DELEGATION_INSTRUCTIONS	Provides concrete delegation strategies with examples: simple queries use 1 sub-agent, comparisons use 1 per element, multi-faceted research uses 1 per aspect. Sets limits on parallel execution (max 3 concurrent) and iteration rounds (max 3).
RESEARCHER_INSTRUCTIONS	Guides individual research sub-agents to conduct focused web searches. Includes hard limits (2-3 searches for simple queries, max 5 for complex), emphasizes using think_tool after each search for strategic reflection, and defines stopping criteria.

Custom Tools

The deep research agent adds the following custom tools beyond the built-in deepagent tools. You can also use your own tools, including via MCP servers. See the Deep Agents package README for more details.

Tool Name	Description
tavily_search	Web search tool that uses Tavily purely as a URL discovery engine. Performs searches using Tavily API to find relevant URLs, fetches full webpage content via HTTP with proper User-Agent headers (avoiding 403 errors), converts HTML to markdown, and returns the complete content without summarization to preserve all information for the agent’s analysis. Works with both Claude and Gemini models.
think_tool	Strategic reflection mechanism that helps the agent pause and assess progress between searches, analyze findings, identify gaps, and plan next steps.