Convert Any Website into an Agentic AI App

Today’s top AI Highlights:

1. Microsoft’s open protocol to convert any website into an agentic AI app

2. AI workspace agents for deep research, docs, and visual reports

3. Transform any text LLM into a voice assistant with near-real-time responses

4. Vibe code and deploy AI apps directly from Google AI Studio

5. LLM-agnostic alternative to OpenAI Codex

& so much more!

Read time: 3 mins

AI Tutorial

Picture this: you're deep in a coding session when you need to update your project documentation in Notion. Instead of context-switching to a browser, navigating through pages, and manually editing content, you simply type "Add deployment notes to the API docs" in your terminal. The magic happens instantly—your Notion page updates without you ever leaving your development environment.

In this tutorial, we'll build a Terminal-based Notion Agent using MCP and Agno framework. This agent will allow you to interact with your Notion pages through natural language commands directly from your terminal, enabling operations like content updates, searches, block creation, and comment addition.

We share hands-on tutorials like this every week, designed to help you stay ahead in the world of AI. If you're serious about leveling up your AI skills and staying ahead of the curve, subscribe now and be the first to access our latest tutorials.

Don’t forget to share this newsletter on your social channels and tag Unwind AI (X, LinkedIn, Threads) to support us!

Latest Developments

Multi-Agent Platform for All Your Research and Office Tasks 📊🔎📑

This felt like the future of workspace automation, essentially vibe working with AI agents. If you’re juggling between Microsoft or Google Docs for documents, PowerPoint for presentations, Excel and Sheets for spreadsheets, ChatGPT for general answers, and a gazillion tools for generating content, you might want to check out this one.

Skywork Super Agents, a multi-agent platform of task-specific automation that can generate professional documents, slides, spreadsheets, websites, podcasts, and more from a simple prompt. Each agent is built for real use cases like business research, data analysis, content creation, and multimodal generation. The platform's standout feature is its Deep Research capability that browses the web in real-time, synthesizes information from multiple sources, and provides clickable citations for every claim

The company has also opensourced its Deep Research Agent framework and released 3 MCP servers for document, data analysis, and presentation, giving developers access to the same AI capabilities powering their platform.

Key Highlights:

1. Deep Research Engine with Source Traceability - Skywork's core differentiator is its deep research capability that performs real-time web searches across multiple sources, synthesizing information into consulting-grade analysis. Every piece of generated content comes with clickable, traceable sources that link back to original web pages, addressing the AI hallucination problem.

2. Specialized Agents for Office Tasks - The platform deploys five expert agents optimized for specific workplace scenarios: Documents for research reports and business plans, Slides for presentations with dynamic visuals, Sheets for data analysis and chart generation, plus Webpages and Podcasts agents for digital content creation. Each agent gathers user requirements before execution, ensuring outputs align with actual needs rather than rushing into blind execution.

3. Opensource Framework and MCP servers - Skywork has opensourced its Deep Research Agent framework as a hierarchical multi-agent system with a top-level planning agent coordinating specialized lower-level agents for analysis, research, and browser automation. They've also published 3 MCP servers that you can integrate directly into your applications for document generation, spreadsheet analysis, and presentation creation.

4. Personal Knowledge Base with Multi-Format Support - The platform includes a NotebookLM-style personal knowledge base where users can upload PDFs, DOCs, PPTs, audio recordings, and URLs to create custom datasets for content generation.

Create How-to Videos in Seconds with AI

Stop wasting time on repetitive explanations. Guidde’s AI creates stunning video guides in seconds—11x faster.

• Turn boring docs into visual masterpieces

• Save hours with AI-powered automation

• Share or embed your guide anywhere

How it works: Click capture on the browser extension, and Guidde auto-generates step-by-step video guides with visuals, voiceover, and a call to action.

Best part? It’s 100% free.

👉 Download the extension now

Turn Any Website Into an AI App for Users and AI Agents 🌐➡️🤖

Microsoft has released NLWeb, an opensource protocol that makes it easy to turn any website into an AI-powered app. NLWeb gives websites a natural language layer using structured data like Schema.org and RSS, so users and agents can query them like they’d talk to an assistant.

Every NLWeb setup also runs as an MCP server, meaning the same conversational interface works for both human visitors and AI agents browsing the web. The project includes everything from the core protocol to reference implementations, vector database connectors, and UI widgets that sites can drop in to start offering natural language interactions immediately. The system works with your existing models, vector stores, and infra - it's platform-agnostic and simple to deploy.

Key Highlights:

1. Built on Existing Web Standards - Instead of reinventing the wheel, NLWeb uses Schema.org markup and RSS feeds that over 100 million websites already use for SEO and syndication. Most sites can enable conversational interfaces without restructuring their existing data or content management systems. The protocol combines this structured data with LLM capabilities to create rich, contextual responses, going beyond keyword matching.

2. Native MCP Integration - Every NLWeb deployment automatically becomes an MCP server that AI agents can call to query website content. This can help websites participate in the emerging agentic web where AI systems browse and interact with sites autonomously. Publishers maintain control over access and usage policies while getting discovered by AI agents and automated tools.

3. Technology-Agnostic - NLWeb works across all major platforms and lets you pick your preferred stack, supporting everything from OpenAI and Anthropic to local models, plus vector databases like Qdrant and Azure AI Search. The system scales from laptop deployments to cloud clusters.

4. Quick Implementation - The project ships with ready-to-use UI widgets, REST API documentation, and reference code that you can integrate directly into existing apps. Publishers can enable conversational features without building chatbots from scratch or hiring specialized AI teams.

Quick Bites

French AI company Kyutai has released Unmute, a cascaded system that allows you to transform any text LLM into a voice assistant without any fine-tuning, by wrapping it in their speech-to-text and text-to-speech models. You can customize both the AI's personality through prompt engineering and its voice through a 10-second audio sample.
Unmute features streaming capabilities that enable the TTS to begin speaking before the LLM finishes generating its full response, reducing the latency significantly. It also preserves the full capabilities of text LLMs, including function-calling and reasoning. The company plans to open-source both the STT and TTS models in the coming weeks

MongoDB’s Voyage AI has launched two new embedding models: voyage-3.5 and voyage-3.5-lite, offering better retrieval quality than their earlier versions at the same cost. Both models support multiple embedding sizes (2048, 1024, 512, 256) and quantization formats, helping reduce vector DB storage costs by up to 99% in some settings. They outperform OpenAI-v3-large on retrieval tasks across multiple domains, with 8.26% and 6.34% average gains, while using smaller, cheaper embeddings. First 200M tokens are free to try.

Google AI Studio now supports native code editing and app building directly in the browser using Gemini 2.5 Pro. You can generate, edit, and deploy full-stack AI apps with one prompt, get diffs in chat, and ship to Google Cloud Run in a click, no setup needed. Apps use a placeholder API key and run in a secure sandboxed iframe, so you can safely share or fork them without exposing credentials. Native code assist, experimental tools, and new SDK integration make it easier to go from prompt to working web app in minutes.

Prime Intellect has released pi-quant, a fast CPU quantization library that’s over 2x faster than PyTorch’s built-in routines across all tested hardware. It supports parallel de/quantization, multiple datatypes (from float32 to int4), and runtime-optimized kernels for AMD64 and ARM64 architectures. The library comes with a Python API for use with PyTorch, NumPy, or standalone setups.

Tools of the Trade

1. Engine: LLM-agnostic alternative to OpenAI Codex. This remote async AI software engineer connects to your GitHub, GitLab, and task tools like Jira and Linear to pick up issues, write code, run tests, and open PRs. It can use LLMs like Claude 4, Gemini 2.5 Pro, GPT-4o, supports full environment setup, and works asynchronously with your workflow with minimal manual input.

2. MCP-UI: Opensource SDK that lets MCP servers send rich, interactive UI components like raw HTML or external web apps directly as part of the model’s response. The client SDK renders these components and wires up user interactions (like button clicks) to the MCP host, enabling event-driven tool calls inside an MCP-based AI workflow.

3. MindsDB: Opensource federated query engine with a built-in MCP server that allows you to connect and query data from 200+ sources, including databases, SaaS applications like Slack and Gmail, and data warehouses, using either SQL or natural language as if they were a single database.

4. Awesome LLM Apps: Build awesome LLM apps with RAG, AI agents, MCP, and more to interact with data sources like GitHub, Gmail, PDFs, and YouTube videos, and automate complex work.

Hot Takes

1. Async remote agents will happen. It's just a question of which tasks and the form factor. If I instruct an agent to complete a task remotely and need to review the final PR in GitHub hours later, I want to be > 90% sure it's correct or user trust will be eroded. ~
   Varun Mohan
   
   

2. I would pay a 2-3x premium to own $500k in OpenAI & Anthropic. These equities are going to go parabolic once they’re public. ~
   Samuel Spitz

That’s all for today! See you tomorrow with more such AI-filled content.

Don’t forget to share this newsletter on your social channels and tag Unwind AI to support us!

Unwind AI - X | LinkedIn | Threads

Awesome LLM Apps | Sponsor Us

PS: We curate this AI newsletter every day for FREE, your support is what keeps us going. If you find value in what you read, share it with at least one, two (or 20) of your friends 😉