Machines create data. Viam gives them intelligence.

A trained model is just a starting point. Viam is the platform with the full stack that production requires — connect any hardware, run inference at the edge, deploy model updates like software, and close the loop from field data back to training. One system, without custom infrastructure.

Capture the data that matters. Skip the pipeline work.

Declare what you want to capture in a JSON config. Viam handles ingestion, edge buffering, and cloud sync, no custom pipeline required.

• Data queues locally during outages and syncs when connectivity returns
• Intelligent triggers from low-confidence detections or sensor thresholds
• Standardized streams across any camera or sensor
• Swap hardware without rewriting data logic

Move from raw data to trained models faster.

Annotation and training live inside the same platform as your fleet. No exports, no third-party handoffs, no broken handoff between capture and training.

• Label data in-platform directly from captured fleet data
• Support for TensorFlow, PyTorch, and ONNX models
• Native training inside the Viam platform
• Models stored and versioned in the Registry as managed assets

Run inference at the edge. React in real time.

Viam’s ML Model Service runs natively on your hardware. If a machine detects a hazard, it reacts without needing a server.

• Local-first architecture for on-device inference without a network connection
• Hardware acceleration on GPU and TPU to prevent application bottlenecks
• Model outputs called through the same SDKs used for motion control
• Perception and action integrated into a single stack

One machine learns. Every machine benefits.

One machine running inference is powerful. A fleet feeding data back into the same model is something else: every job improves the model, and every improvement deploys fleet-wide. That loop only exists when capture, training, and deployment are connected end to end. That’s Viam.

Your data trains your models. Not ours.

• Data sovereignty: Viam never uses customer telemetry for platform-wide training. Export or delete anytime.
• Zero-trust security: Every machine provisioned with a unique identity. No shared secrets, no open ports, no VPNs.

The SY50 is the same hardware, but with Viam's AI reading, interpreting, and acting on sonar data in real time, it's a fundamentally different product — one that helps captains and crews find more fish, burn less fuel, and reduce bycatch.

The infrastructure was already there. Viam added the AI: a computer vision model deployed in 10 days that feeds real-time wait times for concession stands and restrooms into the fan app, helping guests find the shortest lines and get back to the game fast.

Go from perception to action. No custom pipelines in between.

Connect edge inference, OTA model updates, and motion control in one stack. Nothing custom standing between your model and your machine.