¿Qué es Atoms (MGX)?
Atoms — launched in February 2025 as MGX (MetaGPT X) before rebranding — is a no-code platform where you describe a product idea in natural language and a coordinated team of AI agents handles the rest: research, architecture, product planning, engineering, and testing. Named agent roles (a team-lead coordinator, product manager, architect, engineer, and data analyst in the original MGX lineup) divide the work the way a real product team would.
Beyond code generation, Atoms includes a visual editor for design tweaks, a full-stack backend (Atoms Cloud) with authentication, databases, and hosting built in, a "Race Mode" that compares output from multiple underlying AI models side by side, and one-click deployment. Generated projects can be exported to GitHub, so teams aren't locked into the platform for ongoing ownership of the code.
Atoms is built by Foundation Agents, the team behind the open-source MetaGPT and OpenManus projects. It offers a free tier to start, with a credit-based system scaled to task complexity (feature generation, testing, deployment) and yearly billing discounted up to 21% versus paying monthly. The company lists Amazon, Microsoft, Samsung, and eBay among its customers.
Características clave de Atoms (MGX)
- Multi-agent build team: Specialized AI agents handle product planning, architecture, engineering, and testing from one natural-language brief.
- Visual editor: Adjust design and layout directly without regenerating from scratch.
- Full-stack backend included: Atoms Cloud provides authentication, databases, and hosting without separate setup.
- Race Mode: Compare output from multiple AI models on the same task side by side.
- Code export to GitHub: Take the generated codebase out of the platform for full ownership.
- One-click deployment: Ship the built app without a separate deploy pipeline.
Casos de uso de Atoms (MGX)
- Turn a product idea into a working app without writing code
- Prototype a full-stack tool with built-in auth and a database in one step
- Compare how different AI models handle the same build task via Race Mode
- Export a generated app to GitHub to continue development independently