Custom agent logic in LangChain - Interactive Code Practice

Practice - 5 Tasks

1fill in blank

easy

Complete the code to import the base class for creating a custom agent in LangChain.

LangChain

from langchain.agents import [1]

Drag options to blanks, or click blank then click option'

ABaseSingleActionAgent

BAgentOutputParser

CAgentExecutor

DAgentToolkit

Attempts:

3 left

💡 Hint

Common Mistakes

Importing AgentExecutor instead of the base agent class

Confusing output parser with agent base class

The BaseSingleActionAgent is the base class used to create custom agents with specific logic.

2fill in blank

medium

Complete the method signature to define the custom agent's decision logic.

LangChain

def plan(self, intermediate_steps, [1]):

Drag options to blanks, or click blank then click option'

Aoutputs

Binputs

Cactions

Dcontext

Attempts:

3 left

💡 Hint

Common Mistakes

Using 'outputs' instead of 'inputs'

Using unrelated parameter names

The plan method receives intermediate_steps and inputs to decide the next action.

3fill in blank

hard

Fix the error in the code to correctly return the agent's action.

LangChain

return [1](tool='search', tool_input=inputs['query'])

Drag options to blanks, or click blank then click option'

AAgentResponse

BAgentOutput

CAgentStep

DAgentAction

Attempts:

3 left

💡 Hint

Common Mistakes

Using AgentOutput which is for final output

Using non-existent classes like AgentStep

The AgentAction class is used to specify the tool and input for the agent's next step.

4fill in blank

hard

Fill both blanks to create a dictionary comprehension that maps tool names to their descriptions.

LangChain

tool_descriptions = {tool.[1]: tool.[2] for tool in tools}

Drag options to blanks, or click blank then click option'

Aname

Bdescription

Ctool_name

Ddesc

Attempts:

3 left

💡 Hint

Common Mistakes

Using incorrect attribute names like 'tool_name' or 'desc'

Swapping keys and values

Each tool has a name and a description attribute used here to build the dictionary.

5fill in blank

hard

Fill all three blanks to define a custom agent class with a name, a method, and a return statement.

LangChain

class CustomAgent([1]):
    def [2](self, intermediate_steps, inputs):
        return f"Hello, [3]!"

Drag options to blanks, or click blank then click option'

ABaseSingleActionAgent

Bgreet

Cinputs['name']

Dplan

Attempts:

3 left

💡 Hint

Common Mistakes

Using a method name other than 'plan'

Not inheriting from the correct base class

Incorrectly accessing input data

The class inherits from BaseSingleActionAgent. The method to define agent logic is plan. The return uses the input's 'name' key.