Manus AI Builds Its Own Opensource Alternative

Today’s top AI Highlights:

1. Agentic Reasoning - LLMs using AI agents for Deep Research

2. Manus AI created its own open-source alternative in 25 min

3. Chat with any MCP server in your own app

4. Open source collective for building The Internet of Agents

& so much more!

Read time: 3 mins

AI Tutorials

AI Agent Tutorial

There's been quite a debate about whether the Model Context Protocol (MCP) really brings anything new to the table compared to traditional APIs. While some say it just adds another layer of complexity, others point to its standardization benefits. In this tutorial, we'll show you how MCP can actually simplify things when building AI agents that talk to external services like GitHub.

Today, we'll build a GitHub agent that uses MCP to let you query repositories with natural language. You'll track issues, analyze PRs, and check repo activity—all without leaving your chat interface or wrestling with complex API code.

We have used Agno, a lightweight framework for building multi-modal AI agents with a focus on simplicity, performance, and flexibility. With Agno's new MCP integration, you can easily create agents that connect to any MCP-compatible service with minimal code. OpenAI’s GPT-4o is used as the LLM.

AI Workflow

This workflow combines Grok-3's image generation capabilities with Pika AI's video animation features to create stunning transformation videos that show the evolution from vintage to modern aesthetics. Perfect for photo restorations, concept visualizations, or creative storytelling.

We share hands-on tutorials like this 2-3 times a 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, Facebook) to support us!

Latest Developments

LLMs with External Tool-Using AI Agents for Reasoning 🌐🧑‍💻🧠

The University of Oxford researchers have released Agentic Reasoning, a new framework that significantly boosts the reasoning capabilities of LLMs. This approach moves beyond relying on internal knowledge alone combined with Chain of Thoughts. LLMs engage with specialized, external AI agents to actively delegate tasks and solve complex problems.

These agents handle web searches, code execution, and build dynamic knowledge graphs to track complex logical relationships. This allows the main LLM to orchestrate multi-step reasoning processes, tackling problems that require deep research and analysis.

Key Highlights:

1. Agent-Based Architecture - Agentic Reasoning uses external AI agents as tools. The core agents are for web searching (information retrieval), code execution (computational tasks), and creating Mind Maps (structured knowledge representation for maintaining logical flow). You can even swap in, customize, or extend these agents for specific tasks.

2. PhD-Level Performance - The framework was tested on the GPQA dataset, a challenging benchmark of PhD-level science questions. Agentic Reasoning achieved high accuracy rates in chemistry, physics, and biology, rivaling top-tier closed-source models.

3. Improved Explainability - Unlike "black box" reasoning models, Agentic Reasoning's reliance on a Mind Map agent and explicit task delegation to other agents makes the reasoning process more transparent. You can trace the steps and understand how the model arrived at a solution.

4. Beyond Simple RAG - This goes beyond basic retrieval-augmented generation. The LLM actively plans and executes multi-step reasoning strategies, using the agents as tools to dynamically gather information, perform calculations, and maintain logical consistency.

You might think that using external tools or even AI agents with LLMs isn't new, but the specific architecture, the Mind Map agent, the level of orchestration, and the demonstrated performance on complex reasoning tasks are what differentiate Agentic Reasoning from other work.

Manus AI Creates Its Own Open-source Alternative 🔏👨‍🔧

Everyone's buzzing about Manus AI and hunting for invite codes, but here's something even better: Manus AI created its own open-source alternative called ANUS (Autonomous Networked Utility System). In just 25 minutes, this AI agent built a complete framework from scratch, handling everything from design and architecture to code and documentation. All it took was a prompt Manus AI and push to Git using Cursor.

ANUS is a powerful and flexible AI agent framework that aims to deliver similar capabilities as Manus AI. You can use ANUS to create and use AI agents that can execute complex tasks, collaborate in multi-agent environments, interact with various tools, and process multiple types of input.

Key Highlights:

1. Hybrid Agent Architecture - ANUS employs a hybrid agent system that can seamlessly switch between single-agent and multi-agent modes based on task complexity. It has a sophisticated planning system that automatically breaks down complex tasks into manageable steps.

2. Multi-Agent Collaboration - ANUS provides mechanisms for creating societies of specialized agents (like Researchers, Coders, and Planners) that can communicate, reach consensus, and even resolve conflicts. This offers a powerful framework for deploying multi-agent systems for complex, multi-step workflows.

3. Resource Allocation and Memory - ANUS intelligently allocates computational resources based on task requirements. It includes both short-term and long-term memory systems for context retention for multi-turn conversations.

4. Ready-to-Use Tool Ecosystem - ANUS comes with pre-built tools for web automation (using Playwright), document processing, secure code execution (Python sandbox, with more languages supported), and multimodal input (text, images, audio).

5. Extensibility - It's built for extension via plugins, custom tools, and model adapters, so you can tailor it to your specific needs.

6. Model Support - It supports the OpenAI API (like GPT-4o), open-source models (Llama, Mistral), and even local model deployment, offering maximum flexibility and control over costs and privacy.

Quick Bites

AI21 has launched Jamba 1.6, a new open model family designed specifically for enterprise deployment. Jamba 1.6 Large (with 94B active parameters) and Mini (with 12B active parameters) boast top-tier performance that rivals models from Mistral, Cohere, and even Llama models. Featuring a 256K context window and hybrid SSM-Transformer architecture, Jamba 1.6 excels at long-context tasks and RAG. The weights are available to download on Hugging Face.

CopilotKit has released an open-source, web-based client for the Model Context Protocol (MCP), allowing any web application to interact with MCP servers. This generalizes access to MCP beyond existing specialized clients (like Cursor, Windsurf, and Claude's desktop app). The client uses CopilotKit for its UI components and a LangChain-based agent to manage the MCP communication and tool orchestration. You can quickly try it out, just grab a URL of the fully managed MCP servers by Composio.

Cisco, LangChain, and Galileo have launched AGNTCY.org, an open-source collective to build the "Internet of Agents." This initiative aims to create a standardized infrastructure and protocols for AI agents from different frameworks to discover, communicate, and collaborate seamlessly. The project is actively seeking contributors to help define and build this interoperable, multi-agent future.

Tools of the Trade

1. TinyLM: Run models like DeepSeek, Llama3.2, and Nomic Embed directly in the browser with WebGPU acceleration. It offers an OpenAI-compatible API for text generation and embeddings with features like streaming, progress tracking, and TypeScript support.

2. Fast-agent: Python framework for building AI agents that uses MCP servers with minimal code and simple syntax. It lets you create complex agent workflows (chains, parallel execution, evaluation loops) and interact with agents at any stage to improve their performance.

3. Smithery AI: A centralized registry and hosting platform for MCP servers that enables developers to discover, deploy, and integrate LLM extensions following the MCP specification.

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

Hot Takes

1. "Manus is just a wrapper"
   Cursor, Glean, Perplexity, Moveworks, Windsurf are all just wrappers without their own models. Wrappers with $50M+ ARR and unicorn valuations.
   You can build great products and businesses on top of models. ~
   Deedy
   
   

2. if your opinion of manus changed after discovering it's a newersonnet wrapper and not some trained-on-potatoes underground chinese lab leak, you've lost the plot
   idgaf if it's a wrapper. if created value, it deserves my respect. care about capabilities, not architecture ~
   Aidan McLaughlin

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 | Facebook

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 😉