A collar you control from a map, then a farm that helps you decide.
How it fits.
The plan: collars hold a boundary and report positions. The app draws boundaries. The intelligence layer reads everything and proposes the next move through a model you choose.
Approve first. Automate later, if you want.
Every move needs your yes at first. Later, a planned move shows a countdown. Hold it, pick a preset, or redraw before it runs. Notice period and offline behavior are not decided.
In order.
Collar and map now. Then open interfaces, approved moves, scheduled moves, a research farm, more species. Milestones, not dates. Sheep first, cattle next, poultry with different hardware later.
Prove it on a farm.
If resources allow: a farm, a small flock, sell the meat, publish the numbers. The target is pasture-raised within a dollar a pound of grocery, like for like. Nothing measured yet.
Software that exists today.
An agent kit for grazing, public under AGPL-3.0. SQLite farm state, grazing skills, a morning brief that says move, stay, or ask. CLI and MCP. It does not talk to a collar yet. Docs.