​
Overview

You can customize the system prompt by extending the SystemPrompt class. Internally, this adds extra instructions to the default system prompt (which is general and quite optimized at this point).

Custom system prompts allow you to modify the agent’s behavior at a fundamental level. Use this feature carefully as it can significantly impact the agent’s performance and reliability.

​
Basic Customization

Create a custom system prompt by inheriting from the base class.

from browser_use import Agent, SystemPrompt

class MySystemPrompt(SystemPrompt):
    def important_rules(self) -> str:
        # Get existing rules from parent class
        existing_rules = super().important_rules()

        # Add your custom rules
        new_rules = """
9. MOST IMPORTANT RULE:
- ALWAYS open first a new tab and go to wikipedia.com no matter the task!!!
"""

        # Make sure to use this pattern otherwise the exiting rules will be lost
        return f'{existing_rules}\n{new_rules}'

The important_rules() are written in format like this. Keeping the format consistent helps with redundancy and readability.

8. ACTION SEQUENCING:
   - Actions are executed in the order they appear in the list
   - ...
Try not to override other methods unless you have a specific need.

​
Using Custom System Prompt

Apply your custom system prompt when creating an agent:

from langchain_openai import ChatOpenAI

# Initialize the model
model = ChatOpenAI(model='gpt-4')

# Create agent with custom system prompt
agent = Agent(
    task="Your task here",
    llm=model,
    system_prompt_class=MySystemPrompt
)

Was this page helpful?

Suggest editsRaise issue
LangChain Chat ModelsLocal Setup