python

from dotenv import load_dotenv
load_dotenv(".env", override=True)
%load_ext autoreload
%autoreload 2

Deepagent for Research

We will use the deepagents package to create a research agent. When using the deepagents package, it's important to:

Understand the native tools available
Supply task-specific tools
Supply task-specific instructions
Supply task-specific sub-agents

Task-Specific Tools

You can see an overview of the native tools in the deepagents package README as well as the quickstarts README. We'll extend this with two task-specific tools.

Search Tool

There are different search tools that we can use. For example, we can use Tavily to search for relevant URLs, then fetches the full webpage content.

Think Tool

We'll supply a think tool, which is a useful way to help audit agent decision making.

python

from research_agent.tools import tavily_search, think_tool
tools = [tavily_search, think_tool]

Task-Specific Instructions

Next, let's define task specific instructions using a few prompting techniques for agents:

1. Think Like The Agent

What instructions would you give a new work colleague?

Read the question carefully - What specific information does the user need?
Start with broader searches - Use broad, comprehensive queries first
After each search, pause and assess - Do I have enough to answer? What's still missing?
Execute narrower searches as you gather information - Fill in the gaps.

2. Concrete Heuristics (Prevent "Spin-Out" on excessive tool calls)

Use Hard Limits to prevent the research agent from calling tools excessively:

Stop when you can answer confidently - Don't keep searching for perfection.
Give it budgets - Use 2-3 search tool calls for simple queries. Use up to 5 for complex queries.
Limit - Always stop after 5 search tool calls if you cannot find the right source(s).

3. Show your thinking

After each search tool calling, use think_tool to analyze the results:

What key information did I find?
What's missing?
Do I have enough to answer the question comprehensively?
Should I search more or provide my answer?

python

from datetime import datetime
from utils import show_prompt, format_messages
from research_agent.prompts import (
    RESEARCHER_INSTRUCTIONS,
    RESEARCH_WORKFLOW_INSTRUCTIONS,
    SUBAGENT_DELEGATION_INSTRUCTIONS,
)

python

show_prompt(RESEARCHER_INSTRUCTIONS)

Task-Specific Sub-Agents

You can specify custom subagents as a means of context isolation.

Here's well define a sub-agent that can search the web for information.

python

# Get current date
current_date = datetime.now().strftime("%Y-%m-%d")

# Create research sub-agent
research_sub_agent = {
    "name": "research-agent",
    "description": "Delegate research to the sub-agent researcher. Only give this researcher one topic at a time.",
    "system_prompt": RESEARCHER_INSTRUCTIONS.format(date=current_date),
    "tools": [tavily_search, think_tool],
}

Putting it all together

Instructions

Now, we can look at all of our instructions together.

python

# Limits
max_concurrent_research_units = 3
max_researcher_iterations = 3

# Combine orchestrator instructions (RESEARCHER_INSTRUCTIONS only for sub-agents)
INSTRUCTIONS = (
    RESEARCH_WORKFLOW_INSTRUCTIONS
    + "\n\n"
    + "=" * 80
    + "\n\n"
    +  SUBAGENT_DELEGATION_INSTRUCTIONS.format(
        max_concurrent_research_units=max_concurrent_research_units,
        max_researcher_iterations=max_researcher_iterations,
    )
)

show_prompt(INSTRUCTIONS)

Create the agent

Now, we create our deepagent with these components.

python

from IPython.display import Image, display
from deepagents import create_deep_agent
from langchain.chat_models import init_chat_model
from langchain_google_genai import ChatGoogleGenerativeAI

# Model Gemini 3 
model = ChatGoogleGenerativeAI(model="gemini-3-pro-preview")

# Model Claude 4.5
model = init_chat_model(model="anthropic:claude-sonnet-4-5-20250929", temperature=0.0)

python

# Create the agent
agent = create_deep_agent(
      model=model,
      tools=tools, 
      system_prompt=INSTRUCTIONS,
      subagents=[research_sub_agent],
  )
  
# Show the agent
display(Image(agent.get_graph().draw_mermaid_png()))

python

result = agent.invoke(
    {
        "messages": [
            {
                "role": "user",
                "content": "research context engineering approaches used to build AI agents",
            }
        ],
    }, 
)
format_messages(result["messages"])

python

from deepagents.backends.utils import file_data_to_string

# Convert a specific file to string
file_content = file_data_to_string(result["files"]['/final_report.md'])
show_prompt(file_content)

Trace:

https://smith.langchain.com/public/72d23852-4616-4bcc-8d8a-b0d1905c945b/r