MGX (Now Atoms)
The World's First AI Dev Team
4.4•24 reviews•3.1K followers
The World's First AI Dev Team
4.4•24 reviews•3.1K followers
MGX (Now Atoms) is a multi-agent AI platform based on real software SOPs. Chat with the AI team leader, product manager, architect, engineer, and data analyst 24/7 to create websites, blogs, shops, analytics, games, or anything else you can imagine.
TalkMe AI
Great idea, great product. Congrats on the launch!
Llanai
Give me a list of 5 actual products you made that are currently live in the internet.
Elisi : AI-powered Goal Management App
MGX sounds super ambitious — love the concept of turning AI agents into a full dev team that actually follows proper SOPs. Quick question: how much guidance does a user need to provide upfront? Do I have to write out a detailed spec, or can I just throw in a vague idea like “build me a blog for travel stories” and the team figures out the rest?
Also, curious how the AI handles back-and-forth revisions — do the agents keep track of previous changes, or does it feel like starting over each time? Would love to hear if anyone’s already built something cool with MGX!
Atoms
@williamrobertscott
Thanks for your great questions! You've touched on one of our core design principles - making development as effortless as possible!
The beauty of MGX is that you can indeed just share your vision, and let the AI team handle the execution. Simply tell them your goal clearly, and they'll figure out the details, propose features, and handle the implementation.
Here are some examples of what others have built with minimal input:
An artistic portfolio: https://art-portfolio-xn11wn-v3.mgx.world
A sleek link hub: https://linkhub-7rkvvs-v1.mgx.world
A professional portfolio: https://alex-portfolio-yhx5c3-v1.mgx.world
The best part? You can instantly remix any of these projects and customize them with your content. No need to start from scratch!
MGX (Now Atoms)
@williamrobertscott That's an excellent question. We have numerous examples in AppWorld where you can click "view chat" to see the prompts used. They vary tremendously - from one-line requests to detailed PRDs.
Humans are remarkably powerful communicators. Saying "Write a Linux" represents tens of millions of lines of code, while "Make a personal page" might mean hundreds. Yet humans can leverage precise language to accomplish either. We view natural language programming as a process of transforming "ambiguous to precise" instructions, maximizing leverage at each step. A decade ago, using C macros might give you a 10x text leverage (one byte controlling ten). Today, we're seeing 1000x or greater.
However, language contains many ambiguities. "Buy an apple" could refer to fruit or electronics, and involves considerations about price, quantity, weight, origin, logistics, amount, channel, brand, timing, quality control, organic status, expiration date, and countless combinations. AI doesn't yet have human-like preferences - it functions more as a probability-driven stateless function.
I encourage thoroughly exploring MGX alongside us to discover its boundaries and help improve it. Its leverage is incredibly high, with efficiency far exceeding humans for shorter code generation (roughly under 5,000 lines). I'll be publishing a blog post specifically addressing my thoughts on this soon.
Regarding revisions - yes, MGX agents do maintain context across iterations, allowing them to track previous changes rather than starting from scratch each time. This enables continuous refinement and development that builds on prior work.
MGX (Now Atoms)
@williamrobertscott Thank you for your thoughtful questions! As a core developer, I can share our approach to these challenges.
MGX can work with both detailed specs and high-level ideas. For a 'travel blog' request, our PM agent will first generate structured requirements and confirm with you. This ensures alignment while saving your time on detailed specifications.
Regarding iterations, our agents maintain context throughout the development process, tracking changes and design decisions. Each revision builds upon previous work rather than starting fresh.
I'd encourage you to start with a simple concept and see how our agents break it down. What kind of project are you considering? I can provide more specific technical details based on your use case.
MGX (Now Atoms)
@williamrobertscott We’d love for you to dive into MGX and see what’s possible! Here’s a selection of popular templates from App World to get you started: 🎨 Art Portfolio: https://art-portfolio-xn11wn-v3.mgx.world 🎮 Interactive Game: https://flappysnake-v2.mgx.world 📊 Data Dashboard: https://chart-js-dashboard-bbn89v-v1.mgx.world 🌐 Personal Site: https://alex-portfolio-yhx5c3-v1.mgx.world
These examples are just the beginning—you can Remix any template and quickly turn it into your own unique project. Can’t wait to see what you build! Feel free to share your thoughts with us anytime.
MGX (Now Atoms)
@williamrobertscott Thank you for your thoughtful feedback and great questions! 🙌 MGX is designed to be flexible—whether you provide a detailed spec or just a general idea like “build me a blog for travel stories,” the agents can adapt to your input and collaborate to deliver results. For revisions, the AI team keeps track of previous changes, ensuring a seamless back-and-forth process without starting from scratch. It’s all about making the experience intuitive and efficient. We can’t wait to see the amazing projects users create with MGX! 🚀
This looks really interesting, would love to try! I'm very bullish on multi-agent systems. Congrats on the launch and good luck! 👏
How does it compare with current tools like Devin, or IDEs like Cursor?
Atoms
@new_user__1942023a9284e5313298d03
Thanks for asking! MGX takes a fundamentally different approach from Cursor and Devin. While they focus on being AI coding assistants, MGX leverages MetaGPT's multi-agent framework (45k+ stars on GitHub) to simulate a complete development team.
Think of it like this: instead of a single AI assistant, MGX provides specialized AI agents that collaborate like a real team. Each agent has specific expertise and responsibilities, working together seamlessly through structured SOPs.
Our research paper 'MetaGPT: Meta Programming for A Multi-Agent Collaborative Framework' (https://arxiv.org/abs/2308.00352) details this unique approach, which is fully open-source.
Currently, we excel at web development with various templates and Supabase backend integration, while maintaining project context through iterations. Would love to hear what kind of development work you're interested in! 🚀
MGX (Now Atoms)
@new_user__1942023a9284e5313298d03 Thank you for your kind words and interest in MetaGPT X (MGX)! We're excited about its potential too.
MetaGPT X differentiates itself from tools like Devin and Cursor by taking a multi-agent architecture approach to software development. Here's how it compares:
Architecture: Unlike single-agent systems, MGX employs specialized agents (like product managers, architects, engineers, and QA) that work together with distinct roles and goals—similar to how human teams collaborate.
Context Efficiency: Each agent only processes relevant context, significantly reducing token consumption compared to single-agent approaches.
Specialized Expertise: With dedicated agents for specific tasks, MGX delivers higher quality outcomes—architects focus on stable system design, engineers on implementation, QA on testing, etc.
Extensibility: New agents can be easily added to the system without extensive customization.
Independent Evolution: Individual agents can be optimized separately, allowing for targeted improvements.
Would you be interested in trying out MGX? We'd love to hear about your specific use cases!
MGX (Now Atoms)
@new_user__1942023a9284e5313298d03 Thank you so much for the kind words! 🙌 MGX stands out by offering a fully collaborative multi-agent system with personified roles like PMs and architects, making it feel like a real virtual team. While tools like Devin and Cursor are great, MGX focuses on structured teamwork and real-world SOP integration. We’d love to hear your thoughts once you try it out! 🚀
Manna
How does the platform handle complex, multi-step projects? Can the AI team adapt to niche or highly specific requirements? Either way, this feels like a massive leap forward in AI-powered creation. Can’t wait to try it out—thanks for building this!
MGX (Now Atoms)
@desmond_ren1 For complex, multi-step projects, perhaps you should breaks them down into manageable components. This multi-agent approach truly shines in medium-sized projects, particularly those under 5,000 lines of code.
While we're continuously improving our capabilities for larger codebases, MGX can absolutely adapt to niche requirements - that's actually one of its strengths! The modular architecture allows specific agents to be fine-tuned or replaced without disrupting the entire system, making customization more straightforward than with monolithic AI solutions.
We're excited to see what you create with it! Feel free to reach out with any specific use cases you have in mind.
Atoms
@desmond_ren1
Thank you for these excellent questions! MGX handles complex projects through our structured multi-agent collaboration system based on MetaGPT.
Our AI agents work together following industry-standard SOPs, breaking down complex requirements into manageable tasks. For example, when building data-driven applications, Davie handles data analysis while Alex focuses on development - each contributing their specialized expertise.
For specific requirements, our agents can adapt their workflows while maintaining consistent collaboration patterns. You can see this in action here: https://mgx.dev/app/5ru64d
We're continuously improving their capabilities to handle even more sophisticated projects. Your enthusiasm for AI-powered creation really motivates us - can't wait to see what you'll build! 🚀
MGX (Now Atoms)
@desmond_ren1 Thank you for your attention! MGX handles complex projects by breaking down requirements into manageable steps through natural language. You can use the remix operation to save the current valid version and continue incremental development to avoid token consumption due to too long development history. For specific needs, you can start with an existing template and customize it. We'd love to see what you create on mgx.dev!
MGX (Now Atoms)
@desmond_ren1 MGX is designed to handle complex projects! We're constantly improving its ability to adapt to specific needs. Thanks for the feedback!💡
MGX is a bold experiment, attempting to replace costly human labor in software development with AI agents. I’ve used MGX for some time now, but the current version is far from mature—issues like excessive token consumption, a development process that struggles to converge (a problem exacerbated by MGX’s reliance solely on natural language interaction), and its apparent unpreparedness to handle a sizable user base have created friction. Server instability and a fragile payment mechanism drive many users to churn quickly after onboarding. While positioned against Devin, MGX falls short when compared to tools like Cursor and Trae, lacking the polish to compete. That said, for a startup team, MGX has already managed to pique user interest. Many of its current shortcomings will likely be addressed over time. It’s encouraging to see the MetaGPT team’s relentless efforts in this space, and I’m confident they’ll improve. The natural language programming arena is vast enough to accommodate multiple contenders.
Atoms
@bertramray
Thank you for this incredibly detailed and constructive feedback! Your insights are invaluable for our continuous improvement journey.
We've been actively addressing the points you raised through several iterations. Token consumption has been significantly optimized, server infrastructure expanded, and concurrent processing enhanced to provide a more stable experience.
Regarding the comparison with Cursor and Trae - you're right that MGX takes a distinct approach by focusing purely on natural language programming. This is a deliberate choice based on our conviction that natural language will be the future of human-AI collaboration in development. While this path presents unique challenges, it also opens up new possibilities.
We've also introduced practical features to bridge the gap between natural language and traditional coding:
The 'inspect' function enables one-click resource switching and immediate access to element code references for targeted modifications
Our 'editor module' supports both direct code editing and documentation updates, providing flexibility when needed
Your observation about the market being large enough for multiple approaches resonates deeply with us. We're committed to our vision while continuously refining our execution based on valuable feedback like yours.
We'd love to hear more about your experience and specific use cases. Feel free to reach out to our team directly - your insights help shape our development priorities. 🙏
MGX (Now Atoms)
@bertramray Thank you for your continued deep dive into MGX—your insights are incredibly valuable! We're doubling down on refining the development process to improve convergence and boost success rates. Think of it as giving our AI a bit more ‘wisdom’ so it doesn’t burn through tokens like an over-caffeinated coder. ☕💨
To make MGX smarter, we’re working on embedding more project-level development insights into the system—helping it not just execute, but also think and learn like an experienced engineer. At the same time, we're optimizing trajectory self-learning and iteration during the development process, so MGX can adapt and improve dynamically.
We’re also excited to integrate more of our existing academic research into the system to level up performance such as Self-Supervised Prompt Optimization，FACT: Examining the Effectiveness of Iterative Context Rewriting for Multi-fact Retrieval and AFlow: Automating Agentic Workflow Generation.
Appreciate your patience and feedback—it keeps us on our toes and pushing forward!
It is very frustrating, annoying and irritating to see that I cannot finish testing the developed application as I can do in other similar applications, the daily tokens are never reached and I only give one instruction... any way to improve this issue?
Atoms
@moneyrelax
Thank you for sharing your concerns. Let me explain our token system and share some helpful tips:
Our token limits are set at cost price, aligned with the actual API costs from large language models. We've recently updated our system on AppWorld to provide more flexible token management options.
To help you get the most value from your tokens, here are some practical tips:
For conversations, keep them focused and consider using 'remix' when starting a new topic - this can significantly reduce token usage. If an agent makes a mistake, it's often more token-efficient to remix from the point before the error rather than trying to correct it.
For simpler implementations, you can directly @alex instead of engaging the full team. Sometimes agents may have conflicting information in their memory from previous modifications, in which case providing clear correction helps.
You can also explore successful prompt examples through AppWorld -> View Chat to learn effective communication patterns with the agents.
Would you mind sharing your chat history with us? This would help us provide more specific guidance for your use case. Feel free to email us at service@deepwisdom.ai 🙏
MGX (Now Atoms)
@moneyrelax MGX makes it effortless to build web apps for all kinds of scenarios! Check out a few popular demos showcasing different possibilities: 🎨 Art Showcase: https://art-portfolio-xn11wn-v3.mgx.world 🎮 Fun Mini-Game: https://flappysnake-v2.mgx.world 📊 Data Dashboard: https://chart-js-dashboard-bbn89v-v1.mgx.world 🌐 Personal Site: https://alex-portfolio-yhx5c3-v1.mgx.world
And that’s just the start! Your creativity is the real limit—who knows what unique ideas you’ll bring to life next? Because honestly, every fresh idea is next-level awesome. Can’t wait to see your magic at work!