$NUIT: A Web Browser for LLM Agents

New Token | Quick Glance | Nottelabs ($30m FDV) | @nottecore Via @ogandreakiro & @lucas_gdno

DVM View: Notte ( $NUIT ) turns websites into clear maps for AI, simplifying navigation and task automation, reducing errors and costs in web interaction. It's end user is a developer who wants to equip their agent with an ability to navigate the Internet. To us, Notte's technology is unique, and it's founders have a strong history in AI/ML, and product as a wide TAM (all agents). While we don't see direct comps, we believe valuation base could grow towards the Swarms segment range ($50-$150m).

What is the project about? | Structuring the Web for AI Agents with LLM APIs

Notte is a project aimed at transforming the internet into an agent-friendly environment by converting websites into structured, navigable maps described in natural language.

It provides a backend browser API specifically designed for LLM (Language Model) agents to enhance web navigation, data extraction, and task execution.

How does the project work? | Web Navigation Through Text Maps and Credit-Based API

Notte operates by providing an API that translates any webpage into a structured textual description in plain English, detailing both the data available and the actions that can be performed.

When an LLM agent interacts with this API, it can reason based on the structured context provided. For example, if an agent wants to book a flight, Notte processes the LLM's command and executes the action on a cloud-hosted browser session, updating the webpage state accordingly.

This reduces the complexity of dealing with raw HTML or screenshots, making web interactions more efficient and less prone to errors.

While its tech is open sourced, The API operates on a credit-based system, with payments currently through USD via Stripe, with plans to accept $NUIT soon.

The tokenomics of the project indicate long term vision as 10% locked in a smart contract for 6 months, 7.5% with linear vesting over 6 months, 5% allocated for project funding and maintenance, and 2.5% for crypto partnerships and collaborations.

What technology or infrastructure does it use? | Open-Source, Cloud Browsers, and Diverse LLMs with Python SDK

Core Technology: Open-source technology available on GitHub.

API Infrastructure: Managed cloud browser sessions with features like authentication, caching, and action permissions. It supports HTTPS for all requests.

Language Models: Integrates with multiple LLM providers like Cerebras, Anthropic, OpenAI, and Groq, using a llamux configuration for optimal model selection.

Python SDK: Offers a REST API binding for Notte, requiring Python 3.11 or later.

Web Driver: Compatible with any web driver, defaulting to Playwright for browser management.

What problems does it solve? | Minimizing Errors, Improving Efficiency, Streamlining Tasks

Hallucinations and Complexity: By providing structured descriptions instead of raw HTML or screenshots, Notte minimizes hallucinations (incorrect interpretations by LLMs) and reduces the complexity of web navigation for AI agents.

Efficiency: It lowers token usage, reduces latency, and decreases the cost of processing by providing a more digestible input for LLMs to act upon.

Automation: Enables multi-step actions to be executed more efficiently, such as logging out from a website with fewer API calls.

Who are the founders, and what is their vision? | ML Experts with a Vision for an 'Agentic Internet'

The project is being spearheaded by:

Andrea Pinto ( @ogandreakiro ): An ML researcher specializing in LLMs and AI agents, with experience from MIT, ETH Zurich, EPFL, IBM Research, Amazon, and various AI startups.

Lucas Giordano ( @lucas_gdno ): An ML researcher and engineer with experience at Oracle Labs and in growing AI departments in startups. He graduated from EPFL and is also known for his achievements in trail running.

Their vision is to build an "agentic internet" where AI agents can seamlessly navigate and interact with web content, enhancing automation and reducing the cognitive load on LLMs by managing browser interactions.

What is the roadmap or future potential of the project? | Advancing Agent Capabilities with Secrets, OS Integration, and $NUIT Utility

Immediate Plans:

Implementation of secrets management, VPN, and permissions for agents to handle authentication and geolocation requirements.

Expansion to include operating system-level interactions once browser technology stabilizes.

$NUIT integration into payment systems for API usage.

References:

CA: 2eXamy7t3kvKhfV6aJ6Uwe3eh8cuREFcTKs1mFKZpump

X: https://x.com/nottecore

Web: https://notte.cc

Docs: https://docs.notte.cc

Github: https://github.com/nottelabs/notte