Pricing built around production lines, not seats
Every AMI Labs plan includes your machine's world model, the prediction API, and on-premise deployment. Pay for lines and sensors, not for how many engineers log in.
- Typical pilot duration
- 3 weeks
- Time to first prediction
- <48 h
- Setup fee
- None
Pilot
$4,900
/mo
1 production line, 10 sensors
- 90-day historian window per model
- 1 user seat
- OPC-UA connector included
- REST prediction API
- On-premise deployment
- Email support, 2-day SLA
Most Popular
Plant
$16,900
/mo
20 lines, unlimited sensors
- Full historian depth per model
- 5 user seats
- All Pilot features
- Dashboard for all active models
- Automatic retraining on drift
- Priority support, 4-hour SLA
Enterprise
Custom
Multi-site, unlimited lines
- All Plant features
- Multi-site management console
- Direct API access to model internals
- Dedicated research engineer
- Custom SLAs and uptime guarantees
- Procurement and MSA support
Common questions
No. AMI Labs reads from the OPC-UA server or SCADA historian your plant already runs. The connector retrieves existing tag data. You do not need to purchase or install new sensors, and there is no hardware involved in the initial deployment.
For a standard Pilot configuration (10 sensors, 90 days of history), training completes in 4-8 hours on the on-premise node. Larger configurations with full historian depth and more channels take longer, typically overnight. The model is validated before it goes live, so the first prediction you see has already been tested against a held-out window.
The model monitors its own prediction accuracy on a rolling basis. When it detects statistically significant drift, a retraining job is triggered automatically (Plant plan and above). On the Pilot plan, you can trigger retraining manually from the dashboard or via API. Most equipment changes require only incremental retraining, not a full rebuild.
All plans include on-premise deployment. The OPC-UA connector, model trainer, and inference engine all run on a node inside your network. Raw sensor data does not transit to external servers. Only anonymized model performance metrics are reported back to AMI Labs for quality monitoring, and you can disable that in the configuration if required.
Yes. Most customers run the Pilot on a single critical line for the first 3-6 weeks, confirm the prediction accuracy, and then upgrade to the Plant plan. The models trained during the Pilot carry over automatically. There is no migration effort and no new data pull.