Reservations open now. Be one of the first herds on OpenCollar. Reserve now

Virtual fencing
you actually own.

Collars that hold the line, and an agent that decides each morning whether to move it.4 The software is open source.

OpenCollar pricing expected to start at $249. Fully refundable $50 reservation. Subject to change. Full disclaimer.

Should Flock A move today?

Concept of the farm view. Values are made up.
[ 01 / 06 ]How it works

Ground in. Boundary out.

One question for every herd, every morning: move or stay?

Read the ground

Weather, imagery, soil, and forage will come from Alexandria, powered by Firecrawl, alongside where every collar is.

The roadmap

Decide

Stay, move to a specific paddock, or ask you for one more look. Always with reasons.

The docs

Move the line

The new boundary goes to the collars. The move counts once they say they applied it.

The collar
[ .json ]
# run today's decision for one herd
$ openpasture tool run run_decision_cycle \
    --json '{"herd_id":"herd_1"}'
{
  "action": "MOVE",
  "to_land_unit_id": "paddock_home_4",
  "reasoning": [
    "Paddock 4 has rested 38 days.",
    "12 mm of rain expected tomorrow."
  ],
  "confidence": "high",
  "autonomy": "propose"
}

Shapes from the draft collar protocol and the agent kit's decision contract.

[ 02 / 06 ]The collar
2.2 m
First-fix accuracyPrototype 01, open sky
≤ 2.5 m
Accuracy targetV1, open pasture
60+ d
Without sunV1 target, solar top unit
< 2 min
Setup per collarV1 target: scan, strap, done
[ 03 / 06 ]Control

You stay the operator.

It proposes. You decide how much it does on its own.

Autonomy

Three speeds. Approve every move, let it send on a timer unless you hold it, or let it run.

  • Proposewaits for you
  • Timersends unless held
  • Autosends on its own

Reasoning

It shows its work. Every call comes with its inputs, its reasons, and how sure it is.

inputP4P5
residual~4 infull
rest0 d38 d
rain, 24 h12 mm12 mm
low side wetyesno
  • oc_0012applied v4212:30:04
  • oc_0013applied v4212:30:05
  • oc_0014applied v4212:30:05
  • oc_0015received v4212:30:06
+38 collars

Collars

Every boundary acknowledged. Versions only go up. A move counts once the collars say they applied it.

Offline

Fence logic on the collar. It keeps the last good boundary and holds it with no signal. The network only changes where the line is.

Memory

Every day remembered. What it saw, what it decided, what you changed, and what the grass did next. That record is how it learns your ground.

[ 04 / 06 ]Build log

Built in public

Every step, in the log.

One person, a garage, spare time. Some of it even worked.

[ 05 / 06 ]Open source

Agent ready

Plug in your own brain.

openpasture doesn't resell tokens. Bring the model you already pay for, or none at all.

One command. Connect Claude, ChatGPT, or Cursor over MCP, or hand a coding agent the CLI and skills.

Paste this into your coding agent.

Clone github.com/open-pasture/openpasture-agent-kit, install it with pip install -e ., run openpasture skills list, then use the farm-onboarding skill to set up my farm.
Read the docs

Any model, or none. The kit brings the tools, the land data, and the grazing knowledge. The model does the thinking.

  • ClaudeMCP connector
  • ChatGPTMCP connector
  • CursorMCP connector
  • Coding agentsCLI + skills
  • Hermesplugin adapter
  • No modelheuristic fallback
AGPL-3.0. Fork it on GitHub
[ 06 / 06 ]FAQ

FAQ

Questions, answered.

The collar

Can I buy one today?

You can reserve one. A $50 reservation fee per collar holds your place in line. When production is ready, you configure and confirm your order and the fee is applied to it. Until then it's fully refundable. Reserve now.

What will it cost?

Pricing is expected to start at $249 per assembled collar, with the Battery Extension and camera module as add-ons you choose when you configure your order. The price is confirmed before you order. Or build one yourself from the parts list, about $450 in parts today.

Does it need cell signal?

Not to hold the fence. Boundary logic is designed to run on the collar, which keeps its last good boundary. Signal is for new boundaries and position reports.

Which animals?

Sheep first, then cattle. The first live test is Cody's Great Pyrenees: a test of the electronics, not of livestock containment.

Can I use it without openpasture?

That's the point. The collar protocol is being documented so any software that speaks it can run the collars.

The software

What does the agent cost?

The agent kit is free under AGPL-3.0 on your own machine. Your model provider bills you as usual. Optional hosted software may come later.

Is it all open source?

The agent kit is, under AGPL-3.0. Collar designs will be published as they settle. The hardware license isn't decided yet.

Where does the land data come from?

Alexandria, powered by Firecrawl, will provide weather, imagery, soil, and more for your paddocks. Without it, an open-data fallback fills in the weather.

Can I self-host it?

Yes. The kit runs on its own with SQLite. No cloud product required.

Reservations open

Be one of the first herds.

Pricing expected to start at $249 per collar. A $50 reservation holds your place, fully refundable until you confirm your order.

Are you an agent? Start with the skills
[ * ]Disclaimer
  1. 1

    Pricing is an estimate for the assembled OpenCollar (top unit, strap, and ballast module) in US dollars. It excludes add-ons, shipping, taxes, and fees, and may change before orders are confirmed.

  2. 2

    The OpenCollar is a prototype. Specifications are V1 design targets, not measurements, unless marked as measured. Production and delivery timing have not been set and are subject to change.

  3. 3

    The reservation fee is $50 per collar, paid through Stripe. It is fully refundable at any time before you confirm an order, and applied to the purchase price when you do. A reservation is not an order and does not guarantee a price or a delivery date. See the terms.

  4. 4

    Software features such as proposed and scheduled moves describe the design in progress. What works today is on the build log and in the docs.