Platform
One loop from camera to outcome
Cropite is five products on one platform: a shared edge runtime on the machine, one model registry per crop and region, one outcome ledger and one set of agents that plan what happens next.
At a glance
What the platform shares
- Edge runtime
- Jetson Orin / IGX
- Models
- Per crop and region
- Ledger
- Per plant, per pass
- Agents
- Human approval
The loop
See, act, measure, learn
Folira's detections drive Targis. Acreon records the outcome. Synora makes data for what the field rarely shows. Agrona plans the next pass.
Architecture
Six layers, from boom to boardroom
| Layer | Runs on | Job | Product |
|---|---|---|---|
| Edge units | Jetson Orin or IGX on the machine | Per-plant detection at implement speed | Folira |
| Action layer | Same edge unit, implement controller | Targeting policy and nozzle or tool commands | Targis |
| Synthetic data and training | DGX or cloud GPUs | Rare-case data, per-crop fine-tuning | Synora |
| Farm twin | Cloud, Omniverse | Per-plant field history and planning | Acreon |
| Console and agents | Cloud, NIM microservices | Dashboards, prescriptions, approvals | Agrona |
| Feedback loop | All of the above | Outcomes retrain detection and agronomy models | All |
Flywheel
Every acre makes the next one better
1 · More passes
More plants seen
Each pass adds labelled detections, including the ones the model was unsure about.
2 · More outcomes
Known results
Acreon links each action to what happened a week later.
3 · Better models
Retrained per crop
Synora fills gaps; TAO fine-tunes per crop and region.
4 · Less product
Tighter targeting
Better models spray fewer healthy plants and miss fewer weeds.
Deployment
Where each part runs
On the machine
Edge unit
Inference and nozzle control run on the Jetson unit. The pass never waits for a network.
At the farm
Yard sync
When the machine is in Wi-Fi or cellular range, records sync and model updates download.
In the cloud
Twin, training and agents
Cropite cloud by default. Dedicated or regional hosting is planned for larger customers.
Control plane
Models ship like firmware, with a record
- Model registry
- One pack per crop and region
- Over-the-air updates
- Staged, per machine group
- Rollback
- Previous pack stays on the unit
- Approval
- Grower approves updates before a season
- Version on record
- Every action logs its model version
A model change can change what gets sprayed, so Cropite treats it like a firmware release: staged, logged and reversible.
Every record in the ledger carries the model pack version that made the decision.
Security
Your field data stays yours
Ownership
Growers own their data
Field data is used to serve you. Using it to train shared models needs your opt-in.
Protection
Encrypted in transit and at rest
Edge units sign their records; the cloud stores them encrypted. SOC 2 audit is planned.
Access
Roles per farm and client
Service providers see only the fields their clients share with them.
Next step
See the loop on your own field
Start with footage from one pass. We show you the detections, the actions Targis would take and the record Acreon would keep.