Microsoft Unveils NLWeb to Power the Next Generation of the Web

Microsoft has launched NLWeb, an open-source project designed to make it easy for any website to support AI-powered, natural language interaction.

Announced this week, NLWeb—short for Natural Language Web—allows developers and publishers to add conversational interfaces to their websites, enabling users to ask questions or issue commands using everyday language. The system is model-agnostic, works with existing web data formats, and integrates directly into the emerging ecosystem of AI agents through Model Context Protocol (MCP).

Microsoft describes NLWeb as a foundational tool for the "agentic web"—a future where websites don’t just display content but interact with users and agents in intelligent, context-aware ways.

What NLWeb Does

At its core, NLWeb lets publishers turn their websites into intelligent AI applications. It works by combining existing structured data—like Schema.org markup or RSS feeds—with large language models (LLMs) to create natural language interfaces.

These interfaces can be used by human users (via search bars or assistants) or by AI agents, which can access content and interact with it using Model Context Protocol. Every NLWeb deployment functions as an MCP server, making the site discoverable and accessible to other AI-enabled systems—if the publisher opts in.

Microsoft says the goal is to do for AI what HTML did for the web: give every publisher an accessible, standards-based way to participate.

How It Works

NLWeb pulls from semi-structured data sources that most websites already use, such as:

• Schema.org metadata

• RSS feeds

• JSONL or custom structured content

This data is then enhanced with external context from the LLM—such as geographic information or deeper understanding of how pieces of information relate to each other. For example, a restaurant site might provide basic location data, while the AI layer can infer nearby landmarks, identify popular cuisine styles, or suggest similar dining options based on context. This kind of reasoning, often called semantic inference, helps generate more useful and intuitive responses than simply restating raw data.

NLWeb is designed to be:

• Model-agnostic: Supports all major LLMs and vector databases

• Platform-independent: Runs on all major operating systems

• Open and extendable: Fully open-source and customizable via GitHub

Benefits for Publishers

For web publishers, NLWeb offers a direct way to add AI capabilities without giving up control. They can choose their models, manage their data, and decide how and when to interact with AI agents.

Microsoft frames this as part of a broader shift toward the agentic web economy, where AI agents not only answer questions but initiate transactions, make recommendations, or automate tasks across different platforms. NLWeb allows websites to be part of that ecosystem—on their own terms.

According to Microsoft, this could be especially valuable as more search and discovery shifts toward agent-based tools.

Who’s Using It

A small group of early adopters has already begun implementing NLWeb, including:

• Chicago Public Media

• Common Sense Media

• Eventbrite

• Shopify

• Tripadvisor

• Hearst (Delish)

• O’Reilly Media

• Snowflake

• Inception Labs

• Allrecipes / Serious Eats (DDM)

• Vector DB platforms like Milvus and Qdrant

Microsoft says the project is open to publishers of all sizes and categories and encourages contributions from the open-source community.

NLWeb was created by R.V. Guha, now a Technical Fellow at Microsoft and the creator of web standards like RSS, RDF, and Schema.org.

How to Get Started

The full NLWeb framework and documentation are available on GitHub. Included are:

• Lightweight, customizable code that powers the core natural language interface, along with documentation to help developers extend or adapt it to their specific needs.

• Built-in connectors for popular language models and vector databases, plus clear guidance for integrating additional models of your choice.

• Data ingestion tools to help format and load your content—using Schema.org, JSONL, RSS, and other common formats—into your selected vector database for querying.

• A basic web frontend and UI, making it easy to test the system by sending natural language queries and viewing responses in real time.

Developers can customize every layer of the stack and choose how deeply to integrate AI into their site’s user experience.

What This Means

NLWeb could mark a turning point in how websites are built, discovered, and interacted with. By giving publishers the tools to create natural language interfaces—without locking them into specific models or platforms—Microsoft is opening the door to a more open, interoperable AI-powered web.

It also positions websites to participate in the agentic ecosystem now taking shape, where AI agents browse, query, and act on behalf of users. With NLWeb, sites don’t just display content—they become conversational endpoints that agents can understand, navigate, and even transact with.

This has wide-reaching implications:

• For developers, it lowers the barrier to adding AI features to any site—no proprietary platform required.

• For users, it offers more direct, intuitive ways to get answers and complete tasks online, changing how we search.

• For the web itself, it signals a shift from static pages to dynamic, dialog-based interactions—where language becomes a first-class interface layer that AI agents can locate and communicate with.

In the same way HTML unlocked the visual web, NLWeb could help unlock a conversational one—where Microsoft is helping shape a new generation of AI-powered websites that can speak, respond, and interact in real time.

As AI agents begin to shape how information is found and used online, tools like NLWeb may determine which websites remain visible—and which get left behind in a rapidly evolving AI-driven economy.

Editor’s Note: This article was created by Alicia Shapiro, CMO of AiNews.com, with writing, image, and idea-generation support from ChatGPT, an AI assistant. However, the final perspective and editorial choices are solely Alicia Shapiro’s. Special thanks to ChatGPT for assistance with research and editorial support in crafting this article.