langchain/docs/modules/agents/tools/custom_tools.ipynb

Defining Custom Tools

When constructing your own agent, you will need to provide it with a list of Tools that it can use. Besides the actual function that is called, the Tool consists of several components:

• name (str), is required and must be unique within a set of tools provided to an agent
• description (str), is optional but recommended, as it is used by an agent to determine tool use
• return_direct (bool), defaults to False

The function that should be called when the tool is selected should return a single string.

There are two ways to define a tool, we will cover both in the example below.

In [1]:

# Import things that are needed generically
from langchain import LLMMathChain, SerpAPIWrapper
from langchain.agents import AgentType, Tool, initialize_agent, tool
from langchain.chat_models import ChatOpenAI
from langchain.tools import BaseTool

Initialize the LLM to use for the agent.

In [2]:

llm = ChatOpenAI(temperature=0)

Completely New Tools

First, we show how to create completely new tools from scratch.

There are two ways to do this: either by using the Tool dataclass, or by subclassing the BaseTool class.

Tool dataclass

In [3]:

# Load the tool configs that are needed.
search = SerpAPIWrapper()
llm_math_chain = LLMMathChain(llm=llm, verbose=True)
tools = [
    Tool(
        name = "Search",
        func=search.run,
        description="useful for when you need to answer questions about current events"
    ),
    Tool(
        name="Calculator",
        func=llm_math_chain.run,
        description="useful for when you need to answer questions about math"
    )
]

In [4]:

# Construct the agent. We will use the default agent type here.
# See documentation for a full list of options.
agent = initialize_agent(tools, llm, agent=AgentType.ZERO_SHOT_REACT_DESCRIPTION, verbose=True)

In [5]:

agent.run("Who is Leo DiCaprio's girlfriend? What is her current age raised to the 0.43 power?")

Out [5]:

> Entering new AgentExecutor chain...
I need to find out Leo DiCaprio's girlfriend's name and her age
Action: Search
Action Input: "Leo DiCaprio girlfriend"I draw the lime at going to get a Mohawk, though." DiCaprio broke up with girlfriend Camila Morrone, 25, in the summer of 2022, after dating for four years. He's since been linked to another famous supermodel – Gigi Hadid.I need to find out Gigi Hadid's age
Action: Search
Action Input: "Gigi Hadid age"27 yearsI need to calculate her age raised to the 0.43 power
Action: Calculator
Action Input: 27^(0.43)

> Entering new LLMMathChain chain...
27^(0.43)```text
27**(0.43)
```
...numexpr.evaluate("27**(0.43)")...

Answer: 4.125593352125936
> Finished chain.
Answer: 4.125593352125936I now know the final answer
Final Answer: 4.125593352125936

> Finished chain.

'4.125593352125936'

Subclassing the BaseTool class

In [6]:

class CustomSearchTool(BaseTool):
    name = "Search"
    description = "useful for when you need to answer questions about current events"

    def _run(self, query: str) -> str:
        """Use the tool."""
        return search.run(query)
    
    async def _arun(self, query: str) -> str:
        """Use the tool asynchronously."""
        raise NotImplementedError("BingSearchRun does not support async")
    
class CustomCalculatorTool(BaseTool):
    name = "Calculator"
    description = "useful for when you need to answer questions about math"

    def _run(self, query: str) -> str:
        """Use the tool."""
        return llm_math_chain.run(query)
    
    async def _arun(self, query: str) -> str:
        """Use the tool asynchronously."""
        raise NotImplementedError("BingSearchRun does not support async")

In [7]:

tools = [CustomSearchTool(), CustomCalculatorTool()]

In [8]:

agent = initialize_agent(tools, llm, agent=AgentType.ZERO_SHOT_REACT_DESCRIPTION, verbose=True)

In [9]:

agent.run("Who is Leo DiCaprio's girlfriend? What is her current age raised to the 0.43 power?")

Out [9]:

> Entering new AgentExecutor chain...
I need to find out Leo DiCaprio's girlfriend's name and her age
Action: Search
Action Input: "Leo DiCaprio girlfriend"I draw the lime at going to get a Mohawk, though." DiCaprio broke up with girlfriend Camila Morrone, 25, in the summer of 2022, after dating for four years. He's since been linked to another famous supermodel – Gigi Hadid.I now know Leo DiCaprio's girlfriend's name and that he's currently linked to Gigi Hadid. I need to find out Camila Morrone's age.
Action: Search
Action Input: "Camila Morrone age"25 yearsI have Camila Morrone's age. I need to calculate her age raised to the 0.43 power.
Action: Calculator
Action Input: 25^(0.43)

> Entering new LLMMathChain chain...
25^(0.43)```text
25**(0.43)
```
...numexpr.evaluate("25**(0.43)")...

Answer: 3.991298452658078
> Finished chain.
Answer: 3.991298452658078I now know the answer to the original question.
Final Answer: Camila Morrone's current age raised to the 0.43 power is approximately 3.99.

> Finished chain.

"Camila Morrone's current age raised to the 0.43 power is approximately 3.99."

Using the tool decorator

To make it easier to define custom tools, a @tool decorator is provided. This decorator can be used to quickly create a Tool from a simple function. The decorator uses the function name as the tool name by default, but this can be overridden by passing a string as the first argument. Additionally, the decorator will use the function's docstring as the tool's description.

In [10]:

from langchain.agents import tool

@tool
def search_api(query: str) -> str:
    """Searches the API for the query."""
    return f"Results for query {query}"

In [11]:

search_api

Out [11]:

Tool(name='search_api', description='search_api(query: str) -> str - Searches the API for the query.', args_schema=<class 'pydantic.main.ArgsModel'>, return_direct=False, verbose=False, callback_manager=<langchain.callbacks.shared.SharedCallbackManager object at 0x124346f10>, func=<function search_api at 0x16ad6e020>, coroutine=None)

You can also provide arguments like the tool name and whether to return directly.

In [12]:

@tool("search", return_direct=True)
def search_api(query: str) -> str:
    """Searches the API for the query."""
    return "Results"

In [13]:

search_api

Out [13]:

Tool(name='search', description='search(query: str) -> str - Searches the API for the query.', args_schema=<class 'pydantic.main.ArgsModel'>, return_direct=True, verbose=False, callback_manager=<langchain.callbacks.shared.SharedCallbackManager object at 0x124346f10>, func=<function search_api at 0x16ad6d3a0>, coroutine=None)

Modify existing tools

Now, we show how to load existing tools and just modify them. In the example below, we do something really simple and change the Search tool to have the name Google Search.

In [14]:

from langchain.agents import load_tools

In [15]:

tools = load_tools(["serpapi", "llm-math"], llm=llm)

In [16]:

tools[0].name = "Google Search"

In [17]:

agent = initialize_agent(tools, llm, agent=AgentType.ZERO_SHOT_REACT_DESCRIPTION, verbose=True)

In [18]:

agent.run("Who is Leo DiCaprio's girlfriend? What is her current age raised to the 0.43 power?")

Out [18]:

> Entering new AgentExecutor chain...
I need to find out Leo DiCaprio's girlfriend's name and her age.
Action: Google Search
Action Input: "Leo DiCaprio girlfriend"I draw the lime at going to get a Mohawk, though." DiCaprio broke up with girlfriend Camila Morrone, 25, in the summer of 2022, after dating for four years. He's since been linked to another famous supermodel – Gigi Hadid.Now I need to find out Camila Morrone's current age.
Action: Calculator
Action Input: 25^0.43Answer: 3.991298452658078I now know the final answer.
Final Answer: Camila Morrone's current age raised to the 0.43 power is approximately 3.99.

> Finished chain.

"Camila Morrone's current age raised to the 0.43 power is approximately 3.99."

Defining the priorities among Tools

When you made a Custom tool, you may want the Agent to use the custom tool more than normal tools.

For example, you made a custom tool, which gets information on music from your database. When a user wants information on songs, You want the Agent to use the custom tool more than the normal Search tool. But the Agent might prioritize a normal Search tool.

This can be accomplished by adding a statement such as Use this more than the normal search if the question is about Music, like 'who is the singer of yesterday?' or 'what is the most popular song in 2022?' to the description.

An example is below.

In [19]:

# Import things that are needed generically
from langchain.agents import initialize_agent, Tool
from langchain.agents import AgentType
from langchain.llms import OpenAI
from langchain import LLMMathChain, SerpAPIWrapper
search = SerpAPIWrapper()
tools = [
    Tool(
        name = "Search",
        func=search.run,
        description="useful for when you need to answer questions about current events"
    ),
    Tool(
        name="Music Search",
        func=lambda x: "'All I Want For Christmas Is You' by Mariah Carey.", #Mock Function
        description="A Music search engine. Use this more than the normal search if the question is about Music, like 'who is the singer of yesterday?' or 'what is the most popular song in 2022?'",
    )
]

agent = initialize_agent(tools, OpenAI(temperature=0), agent=AgentType.ZERO_SHOT_REACT_DESCRIPTION, verbose=True)

In [20]:

agent.run("what is the most famous song of christmas")

Out [20]:

> Entering new AgentExecutor chain...
 I should use a music search engine to find the answer
Action: Music Search
Action Input: most famous song of christmas'All I Want For Christmas Is You' by Mariah Carey. I now know the final answer
Final Answer: 'All I Want For Christmas Is You' by Mariah Carey.

> Finished chain.

"'All I Want For Christmas Is You' by Mariah Carey."

Using tools to return directly

Often, it can be desirable to have a tool output returned directly to the user, if it’s called. You can do this easily with LangChain by setting the return_direct flag for a tool to be True.

In [21]:

llm_math_chain = LLMMathChain(llm=llm)
tools = [
    Tool(
        name="Calculator",
        func=llm_math_chain.run,
        description="useful for when you need to answer questions about math",
        return_direct=True
    )
]

In [22]:

llm = OpenAI(temperature=0)
agent = initialize_agent(tools, llm, agent=AgentType.ZERO_SHOT_REACT_DESCRIPTION, verbose=True)

In [23]:

agent.run("whats 2**.12")

Out [23]:

> Entering new AgentExecutor chain...
 I need to calculate this
Action: Calculator
Action Input: 2**.12Answer: 1.086734862526058

> Finished chain.

'Answer: 1.086734862526058'

Multi-argument tools

Many functions expect structured inputs. These can also be supported using the Tool decorator or by directly subclassing BaseTool! We have to modify the LLM's OutputParser to map its string output to a dictionary to pass to the action, however.

In [24]:

from typing import Optional, Union

@tool
def custom_search(k: int, query: str, other_arg: Optional[str] = None):
    """The custom search function."""
    return f"Here are the results for the custom search: k={k}, query={query}, other_arg={other_arg}"

In [25]:

import re
from langchain.schema import (
    AgentAction,
    AgentFinish,
)
from langchain.agents import AgentOutputParser

# We will add a custom parser to map the arguments to a dictionary
class CustomOutputParser(AgentOutputParser):
    
    def parse_tool_input(self, action_input: str) -> dict:
        # Regex pattern to match arguments and their values
        pattern = r"(\w+)\s*=\s*(None|\"[^\"]*\"|\d+)"
        matches = re.findall(pattern, action_input)
        
        if not matches:
            raise ValueError(f"Could not parse action input: `{action_input}`")

        # Create a dictionary with the parsed arguments and their values
        parsed_input = {}
        for arg, value in matches:
            if value == "None":
                parsed_value = None
            elif value.isdigit():
                parsed_value = int(value)
            else:
                parsed_value = value.strip('"')
            parsed_input[arg] = parsed_value

        return parsed_input
    
    def parse(self, llm_output: str) -> Union[AgentAction, AgentFinish]:
        # Check if agent should finish
        if "Final Answer:" in llm_output:
            return AgentFinish(
                # Return values is generally always a dictionary with a single `output` key
                # It is not recommended to try anything else at the moment :)
                return_values={"output": llm_output.split("Final Answer:")[-1].strip()},
                log=llm_output,
            )
        # Parse out the action and action input
        regex = r"Action\s*\d*\s*:(.*?)\nAction\s*\d*\s*Input\s*\d*\s*:[\s]*(.*)"
        match = re.search(regex, llm_output, re.DOTALL)
        if not match:
            raise ValueError(f"Could not parse LLM output: `{llm_output}`")
        action = match.group(1).strip()
        action_input = match.group(2)
        tool_input = self.parse_tool_input(action_input)
        # Return the action and action 
        return AgentAction(tool=action, tool_input=tool_input, log=llm_output)

In [26]:

llm = OpenAI(temperature=0)
agent = initialize_agent([custom_search], llm, agent=AgentType.ZERO_SHOT_REACT_DESCRIPTION, verbose=True, agent_kwargs={"output_parser": CustomOutputParser()})

In [27]:

agent.run("Search for me and tell me whatever it says")

Out [27]:

> Entering new AgentExecutor chain...
 I need to use a search function to find the answer
Action: custom_search
Action Input: k=1, query="me"Here are the results for the custom search: k=1, query=me, other_arg=None I now know the final answer
Final Answer: The results of the custom search for k=1, query=me, other_arg=None.

> Finished chain.

'The results of the custom search for k=1, query=me, other_arg=None.'

In [ ]: