Deploy & manage
Register existing Triton servers or create Kubernetes-backed deployments from the UI.
Open source · Built for Kubernetes
Your inference fleet.
One control plane.
Operate NVIDIA Triton on Kubernetes without stitching together separate tools. Deploy, develop, test, tune, and orchestrate from one focused UI.
- Apache 2.0
- Helm deployment
- Your infrastructure
One cockpit for your stack
- KKubernetes
- TNVIDIA Triton
- S3S3-compatible
- AArgo Workflows
- MMLflow
Everything in reach
From model repository to live inference
Keep the engineering detail. Lose the fragmented workflow. Triton Control connects the day-to-day operations around your inference server.
Develop in the browser
Launch per-user, browser-based code-server workspaces directly in your cluster.
Control model repositories
Reuse S3 profiles, browse repositories, inspect model config, and keep artifacts close.
Test & tune
Run inference requests, inspect latency and metrics, then benchmark with Perf Analyzer.
Track experiments
Run Kubernetes-managed MLflow with persistent storage through an authenticated embedded UI.
Orchestrate workflows
Use the embedded Argo Workflows UI and API behind Triton Control's authenticated proxy.
Development without detours
Move from control plane to workspace
Give each user a browser-based development workspace inside Kubernetes. Launch it from Triton Control and keep development close to deployment, model, and inference operations.
- ✓Browser-based code-server environments
- ✓Per-user workspaces inside Kubernetes
- ✓One UI alongside model operations
A real workflow, end to end
Explore the Iris example
Train an Iris classifier with Argo and S3
The included scikit-learn example shows the full path: configure an S3 secret, submit the workflow, train in an ephemeral container, and persist the model plus evaluation results to S3-compatible storage.
01
ConfigureS3 credentials stay in a Kubernetes Secret.
02
RunArgo fetches the training script and executes it.
03
PersistModel, metrics, accuracy, and labels land in S3.
Open by design
Your models. Your cluster. Your control.
Triton Control adds an operational layer around the tools you already use. It does not hide Triton or lock your workflow into a hosted platform.
Deploy where you work
Install with Helm on Kubernetes or evaluate locally with Compose.
Keep the underlying detail
Stay close to Triton, Kubernetes, S3, MLflow, and Argo concepts.
Build in the open
Read the code, follow the roadmap, and contribute under Apache 2.0.
Take command of your inference stack
One open control plane.
Core Triton workflows.
Start with the quickstart, deploy the Helm chart, and bring your first instance online.