This is a different kind of post. Less philosophy, more mud and conduit.
My partner and I are building a thing in the woods behind our house. It’s called Orpheus — an autonomous research station that monitors wildlife, identifies bird species, tracks corvid vocalizations, counts visitors, makes pretty charts, monitors the weather, and eventually talks back.
The hardware: an NVIDIA Jetson Orin NX, four cameras, four research-grade microphones, a hand-built weatherproof enclosure, and a 12-foot mast in Clare County, Michigan. Everything runs locally. No cloud. No phone home.
The theoretical backbone is Karl Friston’s Free Energy Principle — the same framework from my first post. Orpheus isn’t a trail camera. It’s a self-evidencing agent designed to model its environment and eventually act on those models. The roadmap includes interactive puzzles, robotic actuation, and real-time engagement with crows and other animals. Not to command them — to reach something like a communicative equilibrium.
The “Observe” stack is operational. The “Act” stack is intentionally paused — not because it’s hard to build, but because it’s hard to build responsibly.
A white paper is in progress. The platform is going open source. I’ll be presenting Orpheus at conferences later this year. If you’re local to Michigan, or just a fellow bioacoustics or edge ML nerd, say hi.
More soon.