No. 1
Agricultural AI models
Beyond reasons across the farm's records and the open internet. SEED sees. It is a multimodal foundation model pre-trained on more than 100 million field images.
AI research lab for agriculture
The BrainBehind the Barn
Precision AI is building the digital brain that will grow tomorrow's food.
Our first product is Maya, an AI operator a farmer can call, text, or invite to a meeting.
The product
One operator for the whole farm.
Maya is an AI agent built for agriculture, and her interface is a conversation. Call her from the cab. Text her from the bin yard. Invite her to Monday's meeting with your agronomist.
She books the service call, orders the inputs, finds the part, and files the records, working inside the software your farm already runs.
- TelephoneHer own number, hands-free from any cab
- TextSMS, WhatsApp, and most messaging apps
- MeetingsTeams, Google Meet, Zoom
- In the cabOptional rugged tablet, always on
Seven systems, one voice
The average farm runs on seven systems. Maya connects every line.
Financials live in one program, yield maps in another, agronomy in a third. Maya reads, writes, and updates all of them, so a single question can draw on the whole operation.
Place a call
Sample calls using illustrative farm data.
Maya · Farm wireSample call
Maya: Pulled the yield maps, input invoices and grain tickets for all eleven canola fields. Nine cleared cost of production. North Quarter and Section 14 fell short on fertilizer cost per bushel.
Ask once
You say it. Maya gets it done.
Service calls, input orders, parts, drone surveys, satellite maps, trucking, and the weather that rearranges all of it.
Maya's logAn illustrative day in June
- Spray window opens at 6:15 on the home quarter. Wind 9 km/h from the southwest. Crew texted.Weather · Logistics
- Sprayer 2 reports a boom pressure fault. Part located at the dealer. Service call booked for 1:00.Fleet · Parts
- Fungicide order placed with the retailer for the remaining wheat acres. Delivery Thursday.Inputs · Financials
- Joined the agronomy call. Notes and three action items filed to the field records.Agronomy
- Hail reported east of the creek. Drone survey booked for Section 14 and a satellite pass requested.Imagery · Weather
- Truck 3 unloaded at the elevator. Grain ticket matched to the contract and entered in the books.Logistics · Financials
The model behind her
Beyond is trained to say “I don't know.”
Farming punishes a wrong answer. One bad rate on one field can destroy hundreds of thousands of dollars of crop.
Beyond, the reasoning model behind Maya, answers when it is certain and abstains when the evidence falls short. It is built for accuracy, speed, and token efficiency, which matters when a single drone survey holds more data than most models can afford to read.
Inside BeyondMaya · Farm wireField 9
Re: Group 14 rate for kochia. Resistance status cannot be confirmed from records on file.
No rate issued. Survey booked 0700 tomorrow. Agronomist notified.
Answer withheld
certainty too low
53%
More accurate than frontier models on agriculture tasks
Abstention rate
Reading to be posted
Hallucination rate
Reading to be posted
Internal evaluation.
The lab
Three lines of research. One farming brain.
No. 2
Edge computing
The most powerful models that will fit on a tractor. We accelerate them to run in real time on NVIDIA hardware at the field edge.
No. 3
Physical AI
Machines that act on what the model sees: sprayers that treat one plant at a time, survey rigs that map a field in flight, and autopilot research for the ground and the air.
Physical AI
SOLV: Survey Optics at Ludicrous Velocity.
A lightweight edge computer with up to six cameras. It analyzes every frame in flight and streams back the parts that matter.
See SOLVSOLV · Edge survey unit
- Cameras
- Up to 6 inputs
- Compute
- NVIDIA Orin AGX or Thor AGX
- Uplink
- Cellular or Starlink
- Geolocation
- Millimetre-level
Open research
We publish what we learn.
Datasets, models, and tools go out regularly on GitHub and Hugging Face. MCP and embeddings layers for agents are next.
All releases
Pace
A logger with progress reporting for machine learning jobs. Every progress line carries a machine-readable trailer, so a scheduler or dashboard can estimate when a training run will finish.
pip install precisionai-paceMagniData
An open-source explorer for image datasets. Browse images, masks, and overlays, inspect metrics and embeddings, and fly through clusters in 3D.
AgriTune
Train and evaluate agricultural segmentation decoders on frozen features from a remote ViT encoder.
pip install precisionai-agritune
Impact
Less pesticide.
Less fertilizer.
A smaller footprint.
Food systems produce a third of humanity's greenhouse gas emissions.1 Our models treat the plant that needs it and leave the rest of the field clean.
83%
Less pesticide
Achieved in the field
65%
Less fertilizer
Our research target
The thesis
Every machine that grows food will run on a model. We are building it.
Tractors, sprayers, drones, and the farm office are converging on one intelligence. Our research team in Calgary builds it end to end: the reasoning model, the edge model, the compute that carries both into the field, and the agent farmers talk to.
Now taking calls
Put Maya on the line
Tell us about your operation and we will set up a first call.