For the last few weeks, I've been building a new project called Loopy[dot]computer
Loopy is an orchestrator for event-driven agent workflows. Simply put, it's an easy way to define agent workflows that should run when your data changes.
Loopy is open source, agent-neutral, and code-first. You can version control your workflows, use whatever agent you prefer, and host in infra you control.
I've been thinking about the future of agent automations (and loops, much en vogue these past few weeks) for a long time and while my usual method of exploring ideas is to write, it's been pretty exhilarating to be back in the saddle and shipping serious code. It's been a minute.
Give Loopy a try. Or even better, lmk if you want to contribute!
The tldr.
There are different kinds of "loops" that help agents figure out what they should work on.
1. Event-driven: Systems process sensor data (customer feedback, bug reports, observability etc...) in ways that uncover work that agents need to perform in pursuit of well-defined objectives.
2. Explorers: Agents continuously investigate the search space of possible improvements in ways that reflect company goals, strategy, and taste.
3. Sims: Synthetic users navigate the product to unearth potential issues before customers see them.