Microsoft's Open-source AI Agent Framework

Today’s top AI Highlights:

1. Run multiple parallel Claude Code agents in containers

2. Microsoft’s open-source production-ready AI agent framework

3. Google's 100% free technical guide for startups building AI agents

4. Browser Use can now bypass Cloudflare with just 1 parameter

5. Bolt V2 works with Claude Code, Codex, and Gemini agent

& so much more!

Read time: 3 mins

AI Tutorial

Learn OpenAI Agents SDK from zero to production-ready!

We have created a comprehensive crash course that takes you through 11 hands-on tutorials covering everything from basic agent creation to advanced multi-agent workflows using OpenAI Agents SDK.

What you'll learn and build:

• Starter agents with structured outputs using Pydantic

• Tool-integrated agents with custom functions and built-in capabilities

• Multi-agent systems with handoffs and delegation

• Production-ready agents with tracing, guardrails, and sessions

• Voice agents with real-time conversation capabilities

Each tutorial includes working code, interactive web interfaces, and real-world examples.

The course covers the complete agent development lifecycle: orchestration, tool integration, memory management, and deployment strategies.

Everything is 100% open-source.

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

Run Multiple Parallel Claude Code Agents in Containers 🤖🫙🤖🫙

Parallel coding agents usually mean parallel headaches: merge conflicts, shared environments breaking, dependencies getting out of sync.

Sculptor is a desktop app that containerizes each Claude Code agent so you can run as many as you want in parallel without risking your local environment. This app makes it dead simple to spin up an agent for a refactor while another builds a feature while you're debugging something else entirely.

A standout feature is Pairing Mode, which creates a two-way sync between any agent's container and your IDE, letting you test their work in your actual dev environment and commit changes you like while the agent adapts to your edits in real-time. Sessions persist with full context, so you can reopen any past agent conversation without losing plans, code changes, or progress.

Key Highlights:

1. Live IDE Integration - One-click Pairing Mode syncs an agent's container into your local repo with git state intact, so you can test changes immediately in your dev setup while maintaining bidirectional visibility with the agent.

2. Persistent Agent Memory - All sessions save automatically with complete context - conversations, plans, tool calls, and code changes - letting you resume any past session or explore old experiments without re-prompting from scratch.

3. Conflict Resolution System - Automatically detects merge conflicts, lets you cherry-pick the changes you want, and includes beta Suggestions that flag potential issues in agent code before you commit.

4. Availability - Download the app on Mac (Apple Silicon) or Linux, with Intel Mac and Windows support coming soon. Currently supports Claude Code only, GPT-5 integration is on the roadmap. It is absolutely free during beta and uses your Claude subscription.

Microsoft’s Open-Source Framework for Production AI Agents 🤖🏢

Microsoft just dropped an open-source AI framework that merges the experimental flexibility of AutoGen with the enterprise capabilities of Semantic Kernel into one framework.

Microsoft Agent Framework is now in public preview, giving you a unified platform to build multi-agent systems, complete with observability, security, compliance hooks, and long-running durability baked in from day one.

The framework delivers two core capabilities: AI agents that can call tools and MCP servers, and generate responses; and graph-based workflows that connect multiple agents for complex, multi-step tasks with type-based routing, nesting, and human-in-the-loop patterns. It's built on open standards like MCP, A2A, and OpenAPI, ensuring your agents remain portable and vendor-neutral across different frameworks and environments.

You get access to a broad ecosystem of built-in connectors spanning Azure AI Foundry, Microsoft Fabric, SharePoint, Redis, Pinecone, Qdrant, and more.

Key Highlights:

1. Multiple pipeline types - Build systems with any orchestration pattern from Microsoft Research (sequential, concurrent, group chat, handoff, and magentic orchestration) that are production-ready with durability and governance.

2. Pluggable memory - Choose between Redis, Pinecone, Qdrant, Weaviate, Elasticsearch, Postgres, or custom stores for conversational memory. The framework provides the abstraction, while you control the backend for persistent and adaptive memory with hybrid retrieval.

3. VS Code integration - It brings local development, visualization, and debugging of multi-agent workflows directly into VS Code, so you can build and iterate on apps without leaving your editor.

4. OpenTelemetry observability - Instrument and visualize every agent action, tool invocation, and orchestration step with full tracing through Azure AI Foundry dashboards.

5. Shared runtime - The framework will be integrated with Microsoft 365 Agents SDK and Azure AI Foundry Agent Service to create one unified set of abstractions to prototype locally, debug, scale with enterprise-grade hosting, and publish agents across channels.

Quick Bites

Google's 100% free technical guide for startups building AI agents
Google just dropped a comprehensive technical guide for startups building AI agents, covering everything from initial development with ADK, A2A, and MCP to production deployment on Vertex AI. The guide walks through practical implementation details, including agent-to-agent communication, agentic RAG, GraphRAG, and the operational considerations needed to actually ship these systems at scale.

Browser Use can now bypass Cloudflare with just 1 parameter
Browser Use just launched native cloud infrastructure that bypasses anti-bot protections and CAPTCHAs, baked into their library. Just set use_cloud=True in your existing code, add your Browser Use API key, and your agents browse undetected at scale. No CDP configuration or juggling multiple services required.

Replit Connectors to build apps and automations across tools
Replit just launched Connectors, letting you build apps that integrate with Google, Notion, Linear, Salesforce, and 20+ other services without wrestling with OAuth flows or credential management. Log in once to a service, and any app you build on Replit can tap into it. It’s great for quickly spinning up things like automated Jira-to-Confluence sprint reports or anomaly-detecting Google Sheets analyzers.

Claude is now available in Slack with deep workspace access
Claude has landed in Slack with full workspace integration. You can now DM the AI, tag it in threads, or pull it into the assistant panel with web search and tool access. A really interesting feature is the bidirectional connection: Claude can search through your channels, DMs, and files to prep meetings or compile project updates with actual context from your team's conversations.

Bolt V2 works with Claude Code, Codex, and Gemini agent
The OG vibe coding platform Bolt has released its V2 with major feature releases across the board. First, Bolt v2 now integrates Claude Code, OpenAI Codex, and Gemini directly into its platform, giving you access to the best professional AI agents without leaving the interface. These agents come with deep planning built in, so they map out the entire build before writing code, and they debug autonomously -no more getting stuck in error loops. The backend infrastructure (databases, hosting, storage) is already wired up and ready to use, so you don't have to manually configure anything. Once your app is live, built-in analytics give you visibility into user behavior, tracking adoption rates and usage metrics so you can measure how your product is actually being used. It’s available now!

Tools of the Trade

1. ChartDB Agent - AI agent that lets you design or modify database schemas from scratch using simple prompts and view them as ER diagrams; it then exports SQL you can run. The linked diagram shows how tables, fields, and relationships are laid out visually in its interface. Try it out with no sign-up.

2. Awesome GitHub Copilot Customizations - A community-curated collection of prompts, instructions, chat modes, and configurations to supercharge your use of GitHub Copilot across different languages and tasks.

3. CodeLayer - Open-source IDE wrapper around Claude Code that lets you run multiple parallel sessions, add human approval flows, and provides context engineering tools. The project also includes HumanLayer SDK, which provides human-in-the-loop primitives for AI agents that need oversight on high-stakes operations.

4. 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. cs degrees are going to become like english lit degrees.

   you only do it because you’re passionate about the subject, not because there’s a job waiting when you’re done

   ~ Vik
   
   

2. i can’t believe google still hasn’t invested in proper desktop apps. why doesn’t gemini have a clean, native desktop experience?

   browser first thinking made sense in the 2010s, but in 2025 it’s actively hostile to deep usage. claude & chatgpt on desktop “feel” like environments... gemini feels like a webpage. that alone locks it out of becoming a default go-to for me.

   same thing with google calendar... you can’t even run multiple accounts side by side. imagine a desktop app that handled that seamlessly.

   these little things make a massive difference. but at bigco’s, product teams always default to shipping more features, because those are easier to get promoted with. they almost never prioritize polish, infrastructure, or daily delight. this sucks.

   ~ signüll

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 😉