10x Claude with Microsoft's Open-Source Framework

Today’s top AI Highlights:

1. Microsoft's open-source framework force-multiplies Claude Code

2. Google extends A2A with multi-agent squads and hierarchies

3. Gemini 2.5 Computer Use shatters OpenAI and Anthropic CUA

4. Claude Code plugins: commands, agents, MCPs, and hooks

5. One Python library for all your crawling needs

& so much more!

Read time: 3 mins

AI Tutorial

Build an AI Home Renovation Planner Agent using Nano Banana 🍌🍌

Imagine uploading a photo of your outdated kitchen and instantly getting a photorealistic rendering of what it could look like after renovation, complete with budget breakdowns, timelines, and contractor recommendations. That's exactly what we're building today.

In this tutorial, you'll create a sophisticated multi-agent home renovation planner using Google's Agent Development Kit (ADK) and Gemini 2.5 Flash Image (aka Nano Banana).

It analyzes photos of your current space, understands your style preferences from inspiration images, and generates stunning visualizations of your renovated room while keeping your budget in mind.

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

Microsoft's Open-source Framework Force-Multiplies Claude Code 🔋👩‍💻

Microsoft quietly open-sourced something that changes how developers work with AI coding assistants. Amplifier turns tools like Claude Code from helpful sidekicks into actual force multipliers.

Instead of repeatedly explaining context and hand-holding Claude through every task, Amplifier creates a persistent intelligence layer with 20+ specialized AI agents, an automatic RAG pipeline for your docs and notes, and parallel worktree systems that let you test multiple solutions simultaneously, all while keeping everything under Claude’s compact context.

Built on principles of ruthless simplicity and modular design, it's optimized for the reality that AI works best with "right-sized" tasks that fit within context windows.

Also, all of this is portable. Currently, it’s for Claude Code because it’s the best and most widely used. Even when other better tools come, the amplification system remains.

Key Highlights:

1. Specialized Agents - Ships with 20+ expert agents, including zen-architect for design, bug-hunter for debugging, and security-guardian for vulnerability analysis. You can create new specialized agents using the subagent-architect.

2. Knowledge Extraction - Automatically processes your documentation, specs, and notes into a queryable knowledge graph. It extracts ideas with provenance tracking, identifies contradictions worth preserving, and makes accumulated insights instantly accessible to AI assistants.

3. Parallel Worktree Workflow - Amplifier lets you simultaneously explore multiple implementation approaches with isolated branches and environments. Each worktree maintains its own context while accessing shared knowledge, so you can pick winners rather than committing to one path.

4. Conversation Transcript - Captures complete conversation history before compaction events, preserving all context that would otherwise be lost when hitting token limits. Includes search capabilities across past sessions and one-command restoration of full conversation lineage for continuing complex work.

5. Scenario Tool Framework - Vibe coders can create sophisticated tools by describing goals and thinking processes with simple prompts. The system handles brainstorming possibilities, building implementations, and iterative refinement based on feedback.

Turn AI into Your Income Engine

Ready to transform artificial intelligence from a buzzword into your personal revenue generator

HubSpot’s groundbreaking guide "200+ AI-Powered Income Ideas" is your gateway to financial innovation in the digital age.

Inside you'll discover:

• A curated collection of 200+ profitable opportunities spanning content creation, e-commerce, gaming, and emerging digital markets—each vetted for real-world potential

• Step-by-step implementation guides designed for beginners, making AI accessible regardless of your technical background

• Cutting-edge strategies aligned with current market trends, ensuring your ventures stay ahead of the curve

Download your guide today and unlock a future where artificial intelligence powers your success. Your next income stream is waiting.

Get Your Guide

Google Extends A2A with Multi-Agent Squads and Hierarchies 🔀👭

Google just dropped Agent Gateway Protocol (AGP), an extension to A2A that brings enterprise hierarchy to agent networks. While A2A lets agents talk to each other in a flat mesh, AGP organizes them into squads with gateway routers, solving the real problem enterprises face: routing work to the right team based on security policies, cost, and capabilities.

These squads announce what they can do along with their security level and cost ("we handle payroll, require PII clearance, cost $0.10"), then gateways automatically route requests to the cheapest compliant squad.

AGP introduces three core concepts: Capability Announcements that squads broadcast to gateways, Intent Payloads that turn user requests into declarative goals, and routing that maps capabilities to squads. Google has open-sourced the full Python implementation with test scenarios to show exactly how to build policy-compliant multi-agent systems.

Key Highlights:

1. Policy-Based Routing - Gateways filter routes by matching policy constraints before selecting the lowest-cost option. External vendors can't handle sensitive requests even if they're cheaper, ensuring compliance without manual oversight.

2. Squad-Based Organization - Divides agents into autonomous squads (Finance, Engineering, Legal) that mirror enterprise departments. Each squad announces its capabilities to gateway routers, replacing point-to-point integrations with a structured routing layer that scales to hundreds of specialized agents.

3. Intent-Driven Requests - Agents submit intents declaring what they need accomplished and required policies, without knowing which squad handles it. The gateway's routing logic automatically selects compliant squads, decoupling requesters from the underlying agent topology.

4. Standard Error Handling - Defines JSON-RPC error codes for routing failures: route not found, policy violation, or stale routing table. Gateways return structured errors when intents can't be routed safely, giving calling agents actionable feedback instead of silent failures.

Quick Bites

Gemini 2.5 Computer Use shatters OpenAI and Anthropic CUA
Google has released Gemini 2.5 Computer Use model that lets AI agents interact directly with websites and mobile apps by clicking, typing, and scrolling just like humans do. The model runs in an agentic loop: ingesting screenshots and action history, reasoning about the UI state, then firing off function calls for everything from form submission to drag-and-drop operations.

Here’s what makes this interesting. Browserbase evaluated the model independently, where Gemini 2.5 CU didn't just edge out, it shatters other models like OpenAI CUA and Sonnet 4.5 with better performance, ½ the latency, and almost 1/3rd the cost! It is available in public preview via API on Google AI Studio and Vertex AI. You can also try it in a demo environment hosted by Browserbase. Google has open-sourced a full reference agent implementation so you can start building now.

Claude Code plugins: commands, agents, MCPs, and hooks
Anthropic just launched Claude Code Plugins in public beta, letting developers package and share custom slash commands, agents, MCP servers, and hooks through Git repositories. Think of it as a lightweight extension system: you can now install entire development workflows with a single command, and toggle them on and off as needed. To share these customizations, anyone can build and host plugins and create plugin marketplaces. All you need is a git repo with a .claude-plugin/marketplace.json file.

India pilots e-commerce within ChatGPT, Gemini, and Claude
India has kicked off a nationwide pilot where ChatGPT users in India can now order groceries from BigBasket and buy mobile recharges directly in the chat interface, with Gemini and Claude integration coming soon. The pilot uses UPI to complete transactions entirely within the chatbot interface - no app-switching required - while Razorpay handles the merchant layer. The AI platforms won't see payment data, but they get deeper engagement in a market where UPI already processes 20 billion monthly transactions.

Google ADK adds AG-UI support for frontend agent integration
Google ADK now speaks all three major AI protocols - MCP for tools, A2A for agent-to-agent communication, and AG-UI for user interaction. The AG-UI integration lets you embed entire A2A multi-agent meshes directly into React frontends with bi-directional event streaming, generative UI, human-in-the-loop approvals, and synchronized state between your agent network and application. It's essentially bringing multi-agent coordination out of the backend logs and into the user interface where it belongs.

Exa built the fastest search API and the most accurate one too
Exa just dropped version 2.0 of its search API, and the work here is genuinely impressive. They've built two endpoints that play to opposite strengths: Exa Fast delivers sub-350ms latency (30% faster than competitors), and Exa Deep takes a slower, agentic approach, searching, processing, and searching again to maximize result quality. The system runs on a significantly expanded index of tens of billions of webpages that refresh every minute. Benchmarks show Exa Deep scoring 95.1% on SimpleQA against Brave's 82.2%. It's the kind of focused engineering that makes you rethink what's possible with search infrastructure!

Tools of the Trade

1. Crawlee - Build reliable crawlers by handling URL queues, request routing, storage, proxy rotation, retries, and both HTTP and browser crawling under a common interface. You can pick between lightweight HTML crawlers (using BeautifulSoup or Parsel) or full browser rendering with Playwright depending on your target site.

2. Apollo - Free iOS app by Liquid AI that runs AI models entirely on-device without cloud connectivity, offering both Liquid AI's LFM models and other models via OpenRouter API. You can even run an LLM on your computer via LM Studio or Ollama and connect to it with Apollo to host your own private ChatGPT-like mobile app.

3. Compyle - Coding agent built on Claude Code that uses multiple "overwatcher" agents to continuously validate code against your plan and project patterns, stopping to ask clarifying questions instead of making assumptions. It prevents mistakes during development rather than fixing them afterward. 100% free during beta.

4. FleetCode - Open-source desktop app that runs Claude Code, Codex, and other multiple CLI coding agents in parallel using git worktrees for isolation. Unlike Conductor, it keeps the native terminal interface instead of creating a separate UI layer.

5. Awesome LLM Apps - A curated collection of LLM apps with RAG, AI Agents, multi-agent teams, MCP, voice agents, and more. The apps use models from OpenAI, Anthropic, Google, and open-source models like DeepSeek, Qwen, and Llama that you can run locally on your computer.
   (Now accepting GitHub sponsorships)

Hot Takes

1. I'm beginning to suspect that a key skill in working effectively with coding agents is developing an intuition for when you don't need to closely review every line of code they produce. This feels deeply uncomfortable!
   ~ Simon Willison

   
   

2. You can now invest in Nvidia, Intel, AMD, ARM, OpenAI, Mistral, CoreWeave, Nebius, and more with just one ticker: NVDA

   lol

   ~ Teknium (e/λ)

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 😉