Notte: An Open-Source Web Browsing Agent Framework 56

小编 · 2025-05-15

Notte is an open-source web browsing framework that lets developers easily build, deploy, and scale agents through a single API. It manages browser sessions, automates complex web tasks, and lets agents perceive and act on pages via natural language commands. It also offers secure key management. Notte's standout perception layer turns complex web information into concise natural-language descriptions, enabling more efficient and reliable agents. It delivers better task speed, reliability, and evaluation scores than existing solutions.

Notte: An Open-Source Web Browsing Agent Framework

What is Notte?

Notte is an open-source web browsing framework that allows developers to easily build, deploy, and scale agents through a single API. It manages browser sessions, enables agents to automatically complete complex web tasks, and supports perceiving and operating web pages using natural language commands. In addition, it offers secure key management capabilities. What makes Notte special is its perception layer, which can convert complex web page information into concise natural language descriptions, enabling agents to work more efficiently and reliably. It outperforms existing solutions in terms of task time, reliability, and evaluation scores.

Notte: An Open-Source Web Browsing Agent Framework.jpg

Notte Features

Browser Management: Based on Playwright, it handles browser sessions and windows at the underlying level, allowing convenient management of multiple windows and tabs, as well as browser operations such as startup, shutdown, and page navigation.
DOM Processing: Converts web pages into a structured format suitable for agent processing. This makes it easier for agents to understand the content structure of a web page—for example, parsing and categorizing elements such as headings, paragraphs, links, and lists, so that agents can extract information as needed.
Error Handling: Provides a robust error management mechanism for various browser-related scenarios. Whether it's network connection issues, page load failures, or elements not being found, it can effectively capture and handle these issues, ensuring the stability and reliability of agents during web browsing.
Configurable Options: Supports flexible configuration, including headless mode (i.e., a no-interface mode that can run on servers and other environments, saving resources and improving efficiency), user agent (which can disguise as different browsers or devices to access web pages, meeting compatibility requirements of different websites), and proxy settings (facilitating access to web pages through proxy servers, bypassing network restrictions, or achieving privacy protection).
Security Control: Features built-in web security controls and allows custom Chrome browser parameters to enhance browser security and prevent potential security vulnerabilities and attacks.
Web Scripting: Provides web scripting capabilities as well as state-of-the-art scraping endpoints.
Hosted SDK: Provides a hosted SDK that can manage cloud browser sessions and all library features.
API Endpoints: Provides a rich set of API endpoints for session management, page interaction, agent task execution, and more.

Notte Use Cases

Information Extraction and Analysis: Allows agents to automatically browse multiple web pages and extract specific information, such as key events in news articles, research findings in scientific papers, product information on e-commerce sites, and then analyze and summarize it. For example, automatically collecting market data from multiple industry report websites to analyze market trends.
Task Automation: Capable of simulating human operations on web pages to complete repetitive tasks, such as automatically filling out forms, submitting data, and clicking buttons. For example, automatically searching for jobs that match specific criteria on recruitment websites and submitting resumes.
Intelligent Q&A Systems: As part of an intelligent Q&A system, it enables the system to answer user questions by browsing web pages to obtain the latest and most accurate information. When a user asks a question about a specific topic, Notte can help agents find relevant content on web pages and organize answers.

Notte Installation and Usage

Installation Command: Use uv sync --dev --all-extras and uv run pre-commit install for installation and configuration.
Browser Installation: Use patchright to automate browser interactions. You need to install the required browsers, such as uv run patchright install --with-deps chromium.
Environment Variable Configuration: Copy .env.example to .env and fill in the API key. The default uses groq/llama-3.3-70b-versatile, so GROQ_API_KEY needs to be set. You can also set API keys for other LLM providers.
Web Scraping and Data Extraction:

from notte_sdk import NotteClient
notte = NotteClient()
data = notte.scrape(url="https://pump.fun", instructions="get top 5 latest trendy coins on pf, return ticker, name, mcap")