haive-tools

Tool implementations for Haive agents — search, calculators, APIs, code execution, and integrations.

A curated collection of LangChain-compatible tools, ready to drop into any Haive agent. Each tool is wrapped, tested, and exposed via a consistent interface. Use them with ReactAgent, MultiAgent, MemoryAgent, or any agent that takes tools.

---

Why haive-tools?

LangChain has hundreds of tools but they're scattered across packages, have inconsistent interfaces, and many require fiddly setup. haive-tools curates the most useful ones, fixes their quirks, and packages them as a coherent toolkit.

Tool categories

Category	Tools	What they do
🔍 Search	Tavily, Google Search, DuckDuckGo, SerpAPI, Brave Search	Web search with structured results
🧮 Calculators	Python REPL, math expressions, Wolfram Alpha	Calculation and symbolic math
🌐 Web	Fetch URL, scrape HTML, parse JSON	HTTP requests and web scraping
🐍 Code	Python REPL (sandboxed), shell commands	Code execution
📊 Data	CSV/JSON parsers, structured extractors	Data processing
🎮 Fun APIs	Pokémon (PokéBase), Pokémon TCG	API wrappers for testing
📖 Documents	PDF reader, docx parser, web fetcher	Document loading
🗄️ Databases	SQL execution, schema inspection	Database tools

---

Installation

pip install haive-tools

# For Tavily search:
pip install haive-tools[tavily]

# For all extras:
pip install haive-tools[all]

---

Quick Start

Web Search with Tavily

from haive.tools.search import tavily_search_tool
from haive.agents.react.agent import ReactAgent
from haive.core.engine.aug_llm import AugLLMConfig
import os

# Set TAVILY_API_KEY in env
os.environ["TAVILY_API_KEY"] = "tvly-..."

agent = ReactAgent(
    name="researcher",
    engine=AugLLMConfig(
        tools=[tavily_search_tool],
        system_message="Use the search tool to find information.",
    ),
    max_iterations=5,
)

result = agent.run("What are the latest advances in quantum computing in 2025?")

Calculator + Web Search Combo

from haive.tools.search import tavily_search_tool
from haive.tools.calculators import python_calculator
from haive.agents.react.agent import ReactAgent
from haive.core.engine.aug_llm import AugLLMConfig

agent = ReactAgent(
    name="research_assistant",
    engine=AugLLMConfig(
        tools=[tavily_search_tool, python_calculator],
        system_message="Use search for facts and calculator for math.",
    ),
)

result = agent.run("What is the population of Tokyo divided by the area of Japan?")
# Agent: searches for both values, then calculates ratio

Custom Tool Composition

from langchain_core.tools import tool
from haive.tools.search import tavily_search_tool
from haive.tools.web import fetch_url

@tool
def get_weather(city: str) -> str:
    """Get current weather for a city."""
    # ... your custom logic ...
    return f"Sunny, 72°F in {city}"

agent = ReactAgent(
    name="multi_tool",
    engine=AugLLMConfig(tools=[
        tavily_search_tool,
        fetch_url,
        get_weather,
    ]),
)

---

Tool Routing

All tools work with Haive's automatic tool routing system. You can mix:

• LangChain BaseTool (from @tool decorator) → routed to langchain_tool
• Pydantic models with __call__ → routed to pydantic_tool (stateful tools)
• Plain callables → routed to function

from pydantic import BaseModel

class StatefulSearchTool(BaseModel):
    """Search tool with configurable settings."""
    api_key: str
    max_results: int = 5
    region: str = "us"

    def __call__(self, query: str) -> str:
        # Each instance has its own state
        return f"Searching {self.region}: {query}"

# Two instances with different config
us_search = StatefulSearchTool(api_key="...", region="us")
eu_search = StatefulSearchTool(api_key="...", region="eu")

agent = ReactAgent(engine=AugLLMConfig(tools=[us_search, eu_search]))

---

Tool Reference

Search

• tavily_search_tool — Tavily web search (5 results, news + general)
• google_search_tool — Google Custom Search API
• duckduckgo_search_tool — Free DuckDuckGo search
• brave_search_tool — Brave Search API
• serpapi_search_tool — SerpAPI for Google/Bing/etc.

Calculators

• python_calculator — Sandboxed Python REPL for math expressions
• wolfram_alpha_tool — Symbolic math via Wolfram Alpha API

Web

• fetch_url — HTTP GET with auto-decoding
• scrape_url — Scrape HTML with BeautifulSoup
• parse_json — Parse JSON responses

Code

• python_repl — Sandboxed Python execution
• shell_command — Execute shell commands (with safety checks)

Documents

• read_pdf — Extract text from PDF files
• read_docx — Parse Word documents
• web_loader — Load and parse web pages

---

Documentation

📖 Full documentation: https://pr1m8.github.io/haive-tools/

---

Related Packages

Package	Description
haive-core	Foundation: engines, graphs
haive-agents	Production agents
haive-mcp	Dynamic MCP server tools (1,960+)

---

License

MIT © pr1m8