imnyang
638a3d47ce
- Introduced custom output format instructions with example code. - Detailed connection methods for launching and connecting to browsers, including local and remote options. - Provided guidelines for handling sensitive data securely, including best practices and examples. - Documented supported LangChain chat models with setup instructions and environment variable requirements. - Added instructions for customizing the system prompt to control agent behavior.
2.3 KiB
| applyTo |
|---|
| ** |
🧠 General Guidelines for Contributing to browser-use
Browser-Use is an AI agent that autonomously interacts with the web. It takes a user-defined task, navigates web pages using Chromium via Playwright, processes HTML, and repeatedly queries a language model (like gpt-4o) to decide the next action—until the task is completed.
🗂️ File Documentation
When you create a new file:
-
For humans: At the top of the file, include a docstring in natural language explaining:
- What this file does.
- How it fits into the browser-use system.
- If it introduces a new abstraction or replaces an old one.
-
For LLMs/AI: Include structured metadata using standardized comments such as:
# @file purpose: Defines <purpose>
🧰 Development Rules
- ✅ Always use
uvinstead ofpipFor deterministic and fast dependency installs.
uv venv --python 3.11
source .venv/bin/activate
uv sync
-
✅ Use real model names Do not replace
gpt-4owithgpt-4. The modelgpt-4ois a distinct release and supported. -
✅ Type-safe coding Use Pydantic v2 models for all internal action schemas, task inputs/outputs, and controller I/O. This ensures robust validation and LLM-call integrity.
⚙️ Adding New Actions
To add a new action that your browser agent can execute:
from playwright.async_api import Page
from browser_use.core.controller import Controller, ActionResult
controller = Controller()
@controller.registry.action("Search the web for a specific query")
async def search_web(query: str, page: Page):
# Implement your logic here, e.g., query a search engine and return results
result = ...
return ActionResult(extracted_content=result, include_in_memory=True)
Notes:
- Use descriptive names and docstrings for each action.
- Prefer returning
ActionResultwith structured content to help the agent reason better.
🧠 Creating and Running an Agent
To define a task and run a browser-use agent:
from browser_use import Agent
from langchain.chat_models import ChatOpenAI
task = "Find the CEO of OpenAI and return their name"
model = ChatOpenAI(model="gpt-4o")
agent = Agent(task=task, llm=model, controller=controller)
history = await agent.run()