Today we want to briefly explain why we decided to fork two of the projects which some of our co-founders have been contributing to for several years, and why we believe the end result of this decision will be greater opportunity for everyone involved.
In 2019 while working at Fetch.ai one of our co-founders, David Minarsch, co-created the Autonomous Economic Agent (AEA) framework: the world’s first multi-agent system (MAS) framework for building agent-based applications in web3. You can see the first commit here. Initial foundations for this framework were laid while David and his team were building a trading agent game. Thanks to Fetch.ai’s work on multi-agent systems, David and his team were able to expand and develop the framework over the years, culminating in the v1 release in March of 2021.
At this point, David was working part-time as a consultant from his own company Centrality Labs and partnering with Fetch.ai to further develop the AEA framework. With the help of Oaksprout the Tan, he created a site to demonstrate the framework and related research.
In the summer of 2021, David joined forces with Oaksprout the Tan and David Galindo to co-found Valory AG, a new startup that would go on to create Autonolas.
As we started working to build out the Valory software stack we defined a number of high-level goals we aimed to focus on:
Coding velocity: at Valory we grew the team quickly and needed a way to rapidly iterate on the framework, without having to slow down to align with other parties and their specific agendas.
Developer-first vision: at Valory, we believe in a developer-first approach to building software and want to remove any gatekeepers to code creation and dissemination, in particular reliance on centralized component and agent registries.
Architect-first design: as co-creators of the framework, we know it inside out and our vision for it has significantly evolved since founding Valory. We want to bring this vision to its fullest potential.
Forking is not always the right choice for an open-source software project. When deciding whether or not to fork, what matters most is the potential at stake. In order to realize the criteria above (and various other factors outside of our control), it became clear a fork was necessary. In late October of 2021, we decided to fork the AEA repository and pivot all of our further development efforts into the new fork called “open-aea”, which you can see here.
The fork attempts to maintain compatibility with the legacy AEA wherever feasible, but we cannot make any guarantees about compatibility going forward–as we race towards implementing our many ideas, the pace of change and innovation may make this impossible. We have already released three minor versions since the fork and there’s no slowing down in sight.
In late January of 2022, we also decided to fork the ACN (Agent Communication Network) into the “open-acn”. This will give us the freedom to fully implement the original vision of the co-creators of the ACN, as described in the May 2021 whitepaper.
Valory’s Agent Developer Academies give us continuous insight into the usage patterns, pain points and strengths of the open-aea framework. What’s most important for us is to determine what works well for developers, and to then take those insights and put them into use. Our goal is to continue to improve open-aea, making it a solid, highly useful and powerful production framework.
If you would like to join us on this exciting journey building an amazing multi-agent system stack, don’t hesitate to reach out and apply to join the next academy! The Academy is a 4–6 week cohort-based collaborative learning program that teaches skilled Python developers how to use the Valory stack to build their own cutting edge apps and services, all at no cost to participants. You can find more information on the program here.
More broadly, we invite all developers interested in AEAs and multi-agent systems in general, to join us in improving the open-aea framework and its associated tooling.
We are incredibly grateful for the foundations we were able to lay for the field of MAS in crypto while at Fetch.ai, and invite all builders in the Fetch.ai ecosystem to coordinate with us on the co-evolution of both frameworks. This way, new features emerging on the two frameworks can stay compatible, where feasible, contributing to a larger ecosystem of autonomous agents interoperating.
We hope its openness will enable the community to build incredible things. We invite you to join us, to contribute and to build with these incredibly cool technologies — we can’t wait to see what you’ll do with it!
Want to stay up to date with what we’re doing? Give us a follow on Twitter.