This is a real, working demo. Use the steps to register a model, install the agent, run a canary deployment, and watch drift alerts fire — all wired to the actual MLOps.dev backend.
Each step below is interactive and connected to the real MLOps.dev API. When the backend is offline you'll see realistic mock responses so you can still feel the product.
Give the model a name and pick a format. MLOps.dev stores it in the registry with full version history and hardware compatibility metadata.
→ Supports TFLite, ONNX, TensorRT, PyTorch TorchScriptOne command on your edge device. The open-source agent pulls the latest model from the registry, starts local inference, and begins streaming telemetry to the control plane.
→ <8MB binary. Runs on any Linux ARM device with 256MB RAM.Deploy the new model to a slice of your fleet first. MLOps.dev monitors each device's health metrics before opening the rollout to the full fleet.
→ Health gate: if accuracy drops >5% or latency spikes >2× the rollout halts automatically.MLOps.dev monitors your model's input distributions and output confidence in real-time. When drift crosses the configured threshold, it fires an alert — and can trigger an automatic rollback.
→ Threshold: warn at 40%, auto-rollback at 70%You just saw the full MLOps.dev workflow — model registry, agent install, canary deployment, and drift alerting. This is what every team deploying ML to edge devices needs.