Notte provides a full-stack framework for building and deploying LLM-powered web browsing agents, aiming to enhance speed, reduce costs, and improve reliability compared to existing solutions. It targets developers and researchers building agentic internet applications, offering a structured approach to web interaction.

How It Works

Notte introduces a novel "perception layer" that transforms websites into natural language-described structured maps. This abstraction layer significantly reduces the "DOM noise" and cognitive load for LLMs, enabling smaller, faster models to achieve high accuracy. This approach allows agents to parse webpages more efficiently, leading to faster task completion and lower inference costs, even on specialized hardware.

Quick Start & Requirements

• Install: uv venv --python 3.11, uv pip install notte, uv run patchright install --with-deps chromium
• Requirements: Python 3.11, Chromium, GEMINI_API_KEY environment variable.
• Example:

import notte
agi = notte.Agent(reasoning_model="gemini/gemini-2.0-flash", max_steps=5)
agi.run(task="doom scroll cat memes on google images")

• Documentation: https://github.com/nottelabs/notte

Highlighted Details

• Outperforms competitors in speed, cost, and reliability according to open-operator-evals benchmarks.
• Offers a perception layer to map websites into natural language for LLMs.
• Provides a full stack including core browser infrastructure, agents, and web scripting.
• Includes a hosted SDK and API endpoints for scraping and session management.

Maintenance & Community

• Actively developed by the nottelabs-team.
• Citation available for research use.
• Version 1.4.4 as of the README.

Licensing & Compatibility

• Apache 2.0 license.
• Permissive for commercial use and integration with closed-source projects.

Limitations & Caveats

The project lists several upcoming features including captcha resolution, residential proxies, and enhanced security, suggesting current implementations may be incomplete or unstable in these areas.