What Halter has launched
On 19 November 2026, Halter launched an AI assistant called Mav at its research and development farm near Morrinsville in the Waikato. Halter is the Kiwi company behind solar-powered GPS cow collars. Halter's product page calls Mav an "Agriculturally Intelligent Assistant". It works from the data the collars and the Halter app already hold for each farm: grazing, mating, calving, mobs, feed budgets and treatments.
At launch it's open to New Zealand dairy farms on Halter's Pro and Unlimited plans. Rural News Group puts that at roughly 1,200 customers. Halter lists beef farms, Australian farms and US ranches as "coming soon" but hasn't given any dates. Halter also says Mav has already answered more than 100,000 questions during a beta rollout.
For us, the interesting part isn't that a farm company has added a chatbot. It's that Mav can write to the farm system as well as read from it, and Halter has built approval steps and permissions around that. That's a pattern we expect to see in a lot of New Zealand and Australian vertical software over the next year.
How Mav works
According to Halter, Mav runs on "a leading AI model". The company doesn't say which one. Its own farmers, vets and farm operators shape how the assistant reasons, and Halter says it updates Mav every week using feedback from real farms. Farmers open it from a hoof icon on the right-hand side of the farm map in the Halter app.
The product page describes five main jobs:
- Ask. Farmers put questions in plain language and Mav answers from that farm's Halter records. The launch coverage gave examples like how the herd is looking this morning, which cows need attention, and how this season's pasture growth compares with earlier years.
- Take action. Mav can make changes, such as moving cows between mobs, updating mating dates, logging a supplement, adjusting a feed budget or changing a grazing target. Before anything is saved, the farmer reviews the details and approves them.
- Scheduled updates. A one-off question can become a regular report, for example a daily list of cows on heat.
- Uploads. Farmers can add milk dockets, reports and photos, and Mav reads them together with the collar data.
- Sharing. A conversation can go to a staff member, vet or consultant as a read-only link.
Rural News Group described a mating-season example. On a big dairy farm, around 50 cows might come on heat on the same day. Mav can look at each cow's heat history, timing and intensity together and point out the ones that need a closer look. Farmers Weekly reported that Mav can also spot collar faults, suggest round lengths that fit a grazing rotation, and flag health problems.
What early users report
North Otago dairy farmer Nathan McLachlan told Farmers Weekly that Mav sends him a farm summary by text at 7am each day. It covers the previous day's hectares grazed, pasture covers, residuals, heats, matings and calvings. He also said using Mav has made him pay more attention to whether the data going into Halter is accurate.
Andrew Macky runs a 300-cow farm near Te Awamutu and has used Halter for 18 months. He told RNZ that sorting jobs which used to take him 10 to 15 minutes now take about 30 seconds with Mav. "Time is money," he said.
The data behind it
Halter says its collars have produced more than 1.2 trillion data records so far. It describes its anonymised, aggregated dataset as one of the world's largest combined pasture and cattle datasets. The company's figures are:
- about 10,000 cumulative breeding seasons
- about 7 million heat detections
- about 1.2 million health cases
- more than 500 million collar days on cows
Halter has around two million collared cows worldwide. Founder and chief executive Craig Piggott told RNZ that the company's business is selling hardware and services. He said it uses data mainly in aggregate to improve its products and to train Mav. Halter's Mav page also says farm data stays with Halter and isn't sold or handed to outside AI companies to train their models. Conversations stay private unless the farmer shares them.
The archive launched with it
Halter used the same event to open its Agricultural Intelligence Archive. It's a temperature-controlled room inside an old cattle underpass, dug into a hillside on the Morrinsville farm. Halter says it's the first physical archive in the world set up to record pasture farming insights, and that the idea came from seed banks. The archive holds donated historic logbooks from New Zealand dairy farms and records of older farm technology. Halter plans to add verified findings from Mav in physical form over time.
The first publication, Agricultural Intelligence Volume 1, looks at the 2026 calving and pre-mating period. It draws on aggregated data from about 800,000 collared cows on New Zealand dairy farms. It tracks how pasture growth picked up from winter into spring and how nearly 590,000 cows recovered after calving. Five farmers donated their data for case studies. The digital version is free online, and people can see the physical copies on open days. Rural News Group noted that the open-day dates hadn't been confirmed yet.
Limits and costs
Halter is open about the risks. The Mav page says the assistant can get things wrong, "like any AI". It shows its working and won't change anything without approval. For animal health or complicated calls, it tells farmers to talk to their vet or consultant. Halter presents Mav as a tool farmers use together with their advisers. Anyone on a farm can use it, but it follows the same permissions each person already has in the Halter app.
On cost, Mav is free for eligible New Zealand dairy farms during the beta. Halter says it'll give "plenty of notice" if that changes. None of the coverage gave a future price, a date for leaving beta, or any usage caps. Farmers also need to be on the Pro or Unlimited plan in the first place.
Piggott told Farmers Weekly that Halter had talked about the idea for years and had been fully committed to building it for the past year. "It's only going to get smarter," he said. He also said he expects the work of vets and advisers to change for the better, and that vets using Mav could be more proactive.
What builders can take from it
We think Mav is a useful local example of an agent placed on top of a large, well-structured private dataset. Teams building in New Zealand or Australia can borrow a few lessons from it:
- Approve before writing. Every change goes through a review screen first. That keeps a person responsible for real-world actions like moving stock.
- Reuse existing permissions. Mav follows the app's current access rules, so Halter didn't need to build a separate trust model for AI.
- Make answers recurring. Turning a good question into a daily text puts the agent into a farmer's routine without them having to open the app.
- State your data terms clearly. Halter put its no-sale, no-third-party-training promise on the product page, which answers the first thing a cautious customer is likely to ask.
- Bring in domain experts. Vets and farm consultants help shape how Mav reasons, and Halter collects feedback every week. In specialist fields, a model on its own isn't enough.
Halter hasn't released an API for Mav, so outside developers can't build on it directly yet. Halter also hasn't published any accuracy figures, so we'll be watching for independent results once beef and Australian farms get access.
Sources
- Mav, the Agriculturally Intelligent Assistant (Halter)
- Halter's hillside data archive for agricultural intelligence (RNZ)
- Halter launches Mav AI farming assistant (Rural News Group)
- AI set to reshape farming, Halter founder says (Farmers Weekly)
- Halter launches Mav AI tool for farmers (Farmers Weekly)
Build with people working on the same things
Join the Auckland Society for AI: NZ$599 a year with a ForgeBox and the members' projects area. Our Auckland meetups are open to everyone.