For the past six months, we’ve been building an open-source local agent called Eigent, an open-source alternative of Cowork and was #1 on GitHub Trending! It supports BYOK and can help you to organize local files, automate browsers end-to-end. GitHub: https://github.com/eigent-ai/eigent

At the core is CAMEL’s Workforce system, which is inspired by distributing systems: a root node for task planning and coordination, worker nodes for execution, and an asynchronous task channel. It also supports failure tolerance and recursive workers for long-horizon tasks. All of this is open source.

For browser automation, Eigent uses a two-layer architecture:

- a Python layer for agent reasoning and orchestration

- a TypeScript layer (built on Playwright) for native browser control (DOM ops, SoM markers, occlusion handling)

These two layers communicate asynchronously via WebSockets to keep things low-latency and avoid the limits of Python-only automation. This stack is also open source.

Happy to answer questions or take feature requests!

That said, the hardest problems we face today are not in agent design, but in the local desktop runtime itself. Supporting multiple operating systems, versions, and package mirrors has been extremely painful. Our desktop agent installs Python and TypeScript dependencies on first launch, and supporting this reliably across macOS and Windows has been more complex than we initially expected.

After looking into a VM-based approach that uses Apple’s Virtualization framework to run Ubuntu on macOS, we started wondering whether a similar setup could help.

Could this kind of VM-based runtime or something equivalent realistically solve the cross-platform issues across both macOS and Windows?