⚠️ llml.ai is still in Early-Alpha stage and under active development, please use with caution. It is deployed for showcasing and experimenting purposes right now with a glimpse of what the future of llml.ai might look like.

Introducing llml.ai (Early-Alpha version) - created by @felixlyu_1018 (happy to connect!)

Next-Gen AI-powered Web Browser

Next-Gen online AI-powered web browser can help user turn any website into a llm-friendly markdown content or knowledge base with LLM Agents and blazing fast URL reading and links scraping, and html-to-markdown conversion.

🎬 Scenario

Imagine an experience similar to Cursor + Chrome/Safari, where as a user browses a webpage, AI can help the user categorize the entire webpage directory by level, simultaneously creating a webpage knowledge base for the user.

⛳️ Goal

Empowered by various powerful AI models, continuously create a brand new, efficient, and minimal way of reading, learning, scientific research, Vibe Coding, and engineering development.

🌟 Features - An Agentic Web Browser designed and built specifically for AI (direct web Browse, no download, not an APP)

1️⃣ Millisecond-level ultra-fast + free extraction of any webpage content into Markdown format, allowing one-click copying of content to feed into any AI.

2️⃣ Quickly crawls all subpages under the current webpage, and after organizing, de-duplicating, and categorizing them, returns a sitemap 🔗 link map tree 🌲 navigable by large models, similar to llms.txt but more concise, potentially eliminating the need for RAG.

3️⃣ Ability to freely and manually switch to a real-time view of the current page's actual interface, encompassing most of the original webpage's functionalities.

4️⃣ Operations within the built-in browser can also be completely controlled by AI models.

5️⃣ AI can perceive and have context of the webpage content the user is currently browsing, allowing users to directly ask questions or make requests.

6️⃣ Includes free basic mainstream large models as well as advanced large models available via paid subscription.

Demo Tools

llml-links: from llml.ai self-developed API endpoints. Instantly crawl any webpage and generate a structured sitemap or link map tree. This tool helps you visualize and organize all the links and subpages under a given URL, making it easy to understand the structure of complex sites.

How to use: Just ask the chat to get the links for any URL.

llml-read: from llml.ai self-developed API endpoints. Quickly extract and summarize the main content, metadata, and cleaned HTML from any URL. It's designed for fast, reliable reading and conversion of web content into markdown or structured data, perfect for feeding into LLMs or building your personal knowledge base.

How to use: Just ask the chat to read the content for any URL or manually input the URL in the URL input box in the llml.ai in-site Browser that can be opened by clicking the "Open Browser" button in the chat input.

📞 API & MCP

In a near future, all core backend services will be packaged into independent API services for developers to integrate into existing AI Agents or application services, or to build entirely new ones.

🔩 Tech Stack

@nextjs - App Framework @ai-sdk - AI SDK @vercel - Deployment @shadcn - UI Components @supabase - Database

About Core API - Independently researched and developed

It is not reliant on any existing web scraping APIs like 🔥Firecrawl. Instead, our core API is independently researched and developed by myself and still in active alpha stage too. (approx. with speeds potentially 2-10 times faster, yet at prices that could be 1/3 or even lower! ).

The core dedicated and advanced AI-Oriented web scraping API will be released soon for developers to build or integrate into their own applications. Please subscribe to my X for updates.

🔔 FYI - One more thing

Limitations:

The product has just been released and will undergo continuous iterative development.

⚠️ The llml browser currently is not designed to be substituting for a full-featured web browser, and the foundmental technology is not aiming for interacting with web pages like other browser-use based agents, but rather to be a tool for AI Agents to help you better read webpages and get links.

The long-term goal is to create the world's best and most powerful free in-browser browser, including but not limited to: real-time online search, generating images, code, pages, or deployable websites directly based on the current webpage or all webpages collected by the user, or a series of other tasks that Agents can replace.

Imagine you can also prefix any webpage with "llml.ai/" to transform the current webpage into a richly illustrated knowledge base, similar to deepwiki.

Thanks again for your attention!

If you have any questions, please feel free to give me some feedback (form can be found in the preferences dialog) or connect with me on X.