LangChain Hub & hub.pull() Guide: Master ReAct Prompts for AI Agents

What is LangChain Hub?

LangChain Hub is the official repository for sharing and discovering prompts in the LangChain ecosystem. Think of it as npm for prompts—a centralized place where developers can:

The hub.pull() Function

The hub.pull() function downloads a prompt from LangChain Hub into your Python environment:

from langchain import hub

# Basic usage - pulls latest version
prompt = hub.pull("hwchase17/react")

# Pull specific version using commit hash
prompt = hub.pull("hwchase17/react:50442af1")

# Pull your own private prompt (no handle needed)
my_prompt = hub.pull("my-custom-prompt")

# Pull with specific API URL (for self-hosted)
prompt = hub.pull("owner/prompt", api_url="https://your-langsmith.com")

The function accepts an owner_repo_commit string in formats like:

• owner/prompt_name — Latest version
• owner/prompt_name:commit_hash — Specific version
• prompt_name — Your own prompt (if logged in)

Understanding hwchase17/react

hwchase17/react is the official ReAct prompt template created by Harrison Chase (hwchase17), the founder of LangChain. This prompt implements the ReAct (Reasoning + Acting) framework from the influential October 2022 paper by Yao et al.

Related ReAct Prompts

Harrison Chase maintains several ReAct variants for different use cases:

The ReAct Framework Explained

ReAct (Reasoning and Acting) is a prompting paradigm that enables LLMs to solve complex tasks by interleaving reasoning (thinking) with actions (tool use). It was introduced in the paper "ReAct: Synergizing Reasoning and Acting in Language Models" (Yao et al., 2022).

ReAct vs Chain-of-Thought (CoT)

Advanced: Web Scraping Tools for ReAct Agents

Real-world ReAct agents need access to live data. Web scraping tools are essential for AI agents that need to gather information from websites, monitor prices, or collect research data. Here's how to build a production-ready scraping tool for your agent.

The solution is to use mobile proxies that rotate IPs automatically. Mobile IPs from real 4G/5G networks have higher trust scores than datacenter IPs and are much harder to block.

import requests
from langchain.tools import Tool

def scrape_website(url: str) -> str:
    """
    Scrape a website using rotating mobile proxies.
    Replace with your proxy credentials.
    """
    proxies = {
        "http": "http://user:pass@proxy.proxies.sx:port",
        "https": "http://user:pass@proxy.proxies.sx:port"
    }

    headers = {
        "User-Agent": "Mozilla/5.0 (iPhone; CPU iPhone OS 17_0 like Mac OS X)..."
    }

    try:
        response = requests.get(url, proxies=proxies, headers=headers, timeout=30)
        response.raise_for_status()
        return response.text[:5000]  # Truncate for LLM context
    except Exception as e:
        return f"Error scraping {url}: {str(e)}"

# Create the tool
scrape_tool = Tool(
    name="WebScraper",
    func=scrape_website,
    description="Scrapes a website URL and returns the HTML content. Input: full URL"
)

Best Practices for Production ReAct Agents

# Production: pin to specific version
prompt = hub.pull("hwchase17/react:50442af1")

# Development: use latest
prompt = hub.pull("hwchase17/react")

executor = AgentExecutor(
    agent=agent,
    tools=tools,
    handle_parsing_errors=True,  # Retry on malformed output
    max_iterations=10,           # Prevent infinite loops
    early_stopping_method="generate"  # Let LLM decide when to stop
)

import os
os.environ["LANGCHAIN_TRACING_V2"] = "true"
os.environ["LANGCHAIN_API_KEY"] = "your-langsmith-key"
os.environ["LANGCHAIN_PROJECT"] = "my-react-agent"  # Optional

# ❌ Bad: Too vague
Tool(name="Search", description="Searches things")

# ✅ Good: Specific and actionable
Tool(
    name="GoogleSearch",
    description="Search Google for current information. Use for: "
                "news, prices, facts, recent events. "
                "Input: search query string. "
                "Returns: top 5 search results with snippets."
)

from langchain.agents import create_openai_functions_agent

# Alternative: OpenAI Functions agent (simpler tool use)
functions_agent = create_openai_functions_agent(llm, tools, prompt)

# Use ReAct when you need:
# - Explicit reasoning traces for debugging
# - Complex multi-step planning
# - Non-OpenAI models (Llama, Claude, etc.)

Frequently Asked Questions

from langchain import hub
prompt = hub.pull("hwchase17/react")  # Latest
prompt = hub.pull("hwchase17/react:50442af1")  # Specific version

Conclusion

hub.pull("hwchase17/react") is your entry point to building AI agents that can reason through problems and take real-world actions. LangChain Hub makes it easy to discover, share, and version prompts—and the ReAct framework provides a proven pattern for creating agents that think before they act.

Whether you're building research agents, customer support bots, or autonomous data collection systems, understanding hub.pull() and the ReAct framework is essential for modern AI development. Start with the code examples above, explore LangChain Hub for more prompts, and build agents that can truly reason and act.