I’ve been working on Current for a few month and recently open-sourced it.
The central idea is that AI agents should propose actions rather than execute them directly. A separate runtime validates proposals, manages approval and execution, and records the workflow in a hash-chained commit log.
It’s built in Elixir, using OTP for runtime supervision and a functional approach to workflow state transitions.
The project is experimental, but the core workflow is implemented and tested.
I’d be interested in feedback from the Elixir community, particularly on the runtime architecture and design tradeoffs.
