open source · agpl-3.0 · nothing to buy yet — that's the point

pasture is a perennial crop. we're building the machinery to farm it.

ruminant herds guided by radio collars, and an open decision layer that learns your land well enough to plan every day's grazing. the animals harvest and prune; the farmer teaches and steers; the fence stops being the bottleneck.

read-the-manifesto view-on-github

a working thesis with sources — and an honest inventory of what runs today

a farm as the system will see it

Schematic satellite view: paddock grid over forage-index imagery, herd in the current paddock, dashed next paddock, daily plan panel awaiting farmer review

schematic, not product — paddocks over forage-index imagery; yesterday resting, today occupied, tomorrow planned, pending the farmer's sign-off. our prototypes ingest Planet satellite imagery today.

the argument

rotational grazing works. the evidence is decades deep.

move animals to fresh ground daily and utilization climbs from a third of the grass grown to three quarters. same land, same rain — more harvest, more carrying capacity, more carbon in the soil.

+44% more cow-calf pairs on the same acres, rotational vs continuous, in NDSU's long-term comparison.

almost nobody does it. the bottleneck is labor.

daily moves mean daily fence — 15 to 45 minutes per herd, every day, forever. feedlots didn't win because grass failed; they won because grass didn't scale per person.

16% of US cow-calf operations practice intensive rotation, per USDA data. the extension literature names labor as the reason.

the fix exists — locked behind the wrong business model.

virtual-fence collars already steer cattle from a phone. but every vendor sells proprietary hardware with software rented back as firmware: leased collars, closed data, per-head fees forever.

our counter — collars you own and repair, an AGPL decision layer anyone can read, revenue from hosting instead of rent.

the full argument, with sources →

the system

every farm data source in. one daily grazing plan out.

satellite imagery, drone orthomosaics, pasture sensors, and the farmer's own field notes feed a living model of the land. a decision layer turns the model into each day's plan — where the herd goes, how long it stays, what residual it leaves. collars will carry the plan out, and the land's response comes back as tomorrow's data.

the farmer isn't removed from the loop — the fence is. the system proposes every move; the farmer makes the calls that matter and every correction becomes training signal. a grazier's intuition about their own ground is the most valuable dataset on the farm.

how the loop works →

sense

satellite

Planet imagery — prototypes ingest it today

sense

drone

DJI orthomosaics

sense

ground

cameras & sensors

sense

farmer

field notes & photos

↓

model

a living model of the farm

paddocks, forage, recovery, water, herds — runs on whatever data exists

↓

decide

the daily grazing plan

proposed by the system, reviewed and taught by the farmer

↓

act

collars move the herd

the land's response returns as tomorrow's input

↑ the loop closes — grazing response → observation → model → next plan

no fiction

what runs today, exactly.

this site describes a system that does not fully exist yet. we think the honest way to build it is in public, starting with the inventory. future tense on this site is a promise of intent, not a claim of capability — and this ledger stays current.

the full ledger, in context →

state of the project — august 2026

[x]agent kit — open AGPL-3.0 repo: farm domain objects, tools, knowledge ingestion, agent connectors. scaffolding, not a product.
[x]satellite prototypes — Planet imagery ingestion and map visualizations of real ground.
[~]hosted infrastructure — cloud scaffolding exists; not open to users yet.
[ ]no end-to-end product. no farm runs a working loop today.
[ ]no collar hardware. not designed, not prototyped.
[ ]no autonomy. nothing we've built moves an animal.

if this ledger ever describes more than the repos contain, the ledger is wrong — file an issue.

the plan

software first. telemetry second. hardware third.

prove the decision layer with humans executing the moves, then plug in the actuators. no dates — dates on a page like this would be fiction.

the harness now

connect the AI agents farmers already use to farm state, grazing knowledge, and telemetry. the daily loop ships as move / stay / needs-info recommendations, executed by hand.

telemetry next

productionize the sensing layer — Planet pipelines, DJI ingestion, pasture sensors, and existing collar integrations where vendors allow it.

the open collar the campaign

off-the-shelf components, published designs, owner-repairable. a crowdfunded first production run, built in the open, for the farms the incumbents price out.

the far end later

drones for on-demand imagery, deeper satellite partnerships, and eventually upstream capacity of our own.

early access

get in before the campaign

graziers who want to be design partners, engineers who want to build it, and anyone who wants an open alternative to the lock-in playbook. one email when the collar campaign opens — honest progress before that, nothing else.