Evo.ninja, the AI that Evolves in Real-Time
Learn about the new AI agent that can write its own code.
Introducing: evo.ninja, the AI that evolves in real-time
Exciting news from SuperAGI’s Autonomous Agents Hackathon – our project, evo.ninja won first place 🥇!
The hackathon was a 3-day event where 2,300+ hackers used and created AI agents to solve real-world problems. For our project, we created evo.ninja. Evo is an AI agent that's able to write and learn new functionality on-the-fly to try and handle any task.
You can watch our hackathon presentation in the video below.
You can support Evo by giving its Github repo a star. ⭐
In this article we'll talk about: why we built Evo, its unique features, future plans, and the team helping bring Evo to life!
What problem is Evo solving?
Agents currently deal in very simple primitives. For example, they can make HTTP requests and some file operations. But for many real-world tasks, today’s agents may need additional, custom functionality that they weren’t initially built with. Adding these functionalities means the agent’s developer needs to write additional code, then rebuild and redeploy the agent. This approach simply isn’t scalable enough to handle the growing number of agent use-cases.
The evo.ninja agent solves this problem by implementing a novel "learning" architecture. It does this through two mechanisms: writing new scripts, and searching for existing scripts online. Scripts can be run on-demand at runtime, so no rebuilds or redeploys are required!
How does Evo teach itself?
At a high-level, Evo does the following when given a task:
1. Searches locally for pre-written scripts that could complete the task at hand.
2. If nothing is found locally, Evo will search the evo.net for reputable scripts.
3. Finally if nothing is found, Evo will write its own script to complete the task
In this way, Evo teaches itself new capabilities, similar to how Neo from the Matrix could download abilities on-the-fly to take on new challenges!
The future of Evo
We believe that learning should be collaborative, and so our next chapter is to build evo.net, a network of evo.ninja agents and professional programmers that share scripts with each other. The network would be continuously improved by its participants, who would help curate the most effective scripts.
Who are we?
We are contributors to the Polywrap project, residing from all around the world. We have experience in:
Core development on Polywrap, a framework for developing composable, on-demand WebAssembly-based software.
Developing full-stack web2 and web3 software
Working at tech and financial consulting companies
And more!
We’d love to hear your thoughts on our project. Join our Discord and introduce yourself, subscribe to our reddit, follow us on Twitter, and please support us by giving our Github repo a star ⭐.