What are Haystack Agents? The Complete Guide to Tool-Driven NLP with Code Operations

Modern NLP applications often require multi-step thinking, collaboration with external tools, and the ability to effectively adapt to user queries. Haystack Agents, a new feature of the Haystack NLP framework by deepset, are an example of this new wave of advanced NLP capabilities.

Haystack Agents are designed to handle situations that require:

• Complex multi-step reasoning.
• Integration of external tools or APIs.
• Improved retrieval workflow that goes beyond answering simple queries.

This article delves into the Haystack Agents framework, exploring its features, architecture, and real-world applications. To provide actionable information, we will build a QA Agent that uses tools such as a search engine and a calculator.

Why Choose Haystack Agents?

Unlike general-purpose frameworks such as LangChain, Haystack Agents are deeply integrated within the Haystack ecosystem, making them highly effective at specialized tasks such as document retrieval, custom tool integration, and multi-step reasoning. These agents excel at searching large datasets using advanced add-ons, increasing performance by integrating APIs for operations such as calculations or database queries, and dealing with complex queries that require logical capture. Being open source and modular, Haystack Agents integrate seamlessly with popular ML libraries and infrastructure such as Elasticsearch, Hugging Face models, and pre-trained transforms.

Properties of Haystack Agents

Haystack Agents are built using a tool driven structures. Here, tools act as individual modules designed for specific tasks, such as document search, analytics, or API interaction. The agent dynamically decides which tools to use, the order in which they are used, and how to combine their outputs to form a coherent response. The architecture includes key components such as tools, which implement specific action commands that guide the agent's decision-making process. These finders help search for documents within large data sets, and the nodes and pipelines control data processing and workflow planning in Haystack.

Use Case: Building a QA Agent with Search and Calculator Tools

In this tutorial, our QA Agent will do the following:

• Return answers to factual questions from the document store.
• Perform mathematical calculations using a calculator.
• Dynamically combine the results if needed.

Step 1: Enter the Required

Before entering the implementation, make sure that your location is configured:

1. Install Python 3.8 or higher.

2. Install Haystack and all dependencies:

# bash
pip install farm-haystack[all]

3. Launch Elasticsearch, the backbone of our document store:

# bash
docker run -d -p 9200:9200 -e "discovery.type=single-node" docker.elastic.co/elasticsearch/elasticsearch:7.17.1

Step 2: Launch Document Store and Recovery

A document store is a central repository for storing and querying documents, while the retriever finds the relevant documents for a particular query.

# python
from haystack.utils import launch_es
from haystack.nodes import EmbeddingRetriever
from haystack.pipelines import DocumentSearchPipeline
from haystack.document_stores import ElasticsearchDocumentStore

# Launch Elasticsearch
launch_es()

# Initialize Document Store
document_store = ElasticsearchDocumentStore()

# Add documents to the store
docs = [
    {"content": "Albert Einstein was a theoretical physicist who developed the theory of relativity."},
    {"content": "The capital of France is Paris."},
    {"content": "The square root of 16 is 4."}
]
document_store.write_documents(docs)

# Initialize Retriever
retriever = EmbeddingRetriever(
    document_store=document_store,
    embedding_model="sentence-transformers/all-MiniLM-L6-v2",
    use_gpu=True
)

# Update embeddings
document_store.update_embeddings(retriever)

Step 3: Define Tools

Tools are the building blocks of Haystack Agents. Each tool has a specific purpose, such as searching documents or performing calculations.

# python
from haystack.agents.base import Tool

# Search Tool
search_pipeline = DocumentSearchPipeline(retriever)
search_tool = Tool(
    name="Search",
    pipeline_or_node=search_pipeline,
    description="Use this tool for answering factual questions using a document store."
)

# Calculator Tool
def calculate(expression: str) -> str:
    try:
        result = eval(expression)
        return str(result)
    except Exception as e:
        return f"Error in calculation: {e}"

calculator_tool = Tool(
    name="Calculator",
    pipeline_or_node=calculate,
    description="Use this tool to perform mathematical calculations."
)

Step 4: Start the agent

Agents in Haystack are configured with tools and a quick template that explains how they interact with the tools.

# python
from haystack.agents import Agent

# Initialize Agent
agent = Agent(
    tools=[search_tool, calculator_tool],
    prompt_template="Answer questions using the provided tools. Combine results if needed."
)

Step 5: Ask an Agent

Contact the agent by asking natural language questions.

# python
# Factual Question
response = agent.run("Who developed the theory of relativity?")
print("Agent Response:", response)

# Mathematical Calculation
response = agent.run("What is the result of 8 * (2 + 3)?")
print("Agent Response:", response)

# Combined Query
response = agent.run("What is the square root of 16, and who developed it?")
print("Agent Response:", response)

Advanced Features of Haystack Agents

• Custom Tools: Integrate APIs or domain-specific tools to extend functionality (eg, weather APIs, stock market data).
• Fine Tuned Models: Replace the default embedding model with one optimized for special operations.
• Pipe Chains: Use multiple pipelines to process complex queries involving multiple data sources.

In conclusion, Haystack Agents provide a powerful, flexible, and modular framework for building advanced NLP applications that require multi-step thinking and tooling. With their seamless integration into the Haystack ecosystem, these agents excel at tasks such as document retrieval, custom API integration, and logic processing, making them ready to solve complex real-world problems. They are well suited for applications such as customer support bots, which combine document search with external APIs for real-time ticket processing, educational tools that acquire information and perform calculations to answer user questions, and business intelligence solutions that combine data from multiple sources. sources and produce information.

Sources

Also, don't forget to follow us Twitter and join our Telephone station again LinkedIn Grup. Don't forget to join our 65k+ ML SubReddit.

🚨 [Recommended Read] Nebius AI Studio extends with vision models, new language models, embeddings and LoRA ^(Promoted)