Integration guide
Connect to the platform through training containers, DataOps, MLflow, the container registry and the REST API.
This chapter is for external partners and ML engineers who connect to MLOps Central Platform outside the screens.
How to use the screens is in the User guide. Replace the example addresses (http://localhost:10000, https://mlops.example.com) and tenant DEMO with your own values.
What do you want to do?
| I want to… | Read | Main readers |
|---|---|---|
| run my training code from the platform's training wizard | Build a training container | Partners, ML engineers |
| have the platform pull datasets from our data server | DataOps integration | Partners (data providers) |
| record training run on a laptop or in-house server on the platform's experiment screens | Log directly with MLflow | ML engineers |
use the platform registry with docker push/pull | Use the container registry directly | ML engineers, operators |
| automate uploads, queries and notifications with scripts | Automate with the REST API | ML engineers |
Three kinds of credentials
The platform uses a different token for each entry point.
| Credential | Used for | Issued at | Tenant selection |
|---|---|---|---|
| Login session (cookie or JWT) | REST API (/api/v1/…) | POST /auth/cookie/login, POST /auth/jwt/login | X-Tenant header on every request |
| MLflow token | MLflow client (/mlflow) | Account settings (계정 설정) → Manage MLflow tokens (MLflow 토큰 관리), POST /api/v1/mlflow/tokens | Fixed in the token — no header needed |
| Container token | docker login (/v2) | Account settings (계정 설정) → Manage container tokens (컨테이너 토큰 관리), POST /api/v1/registry/tokens | Fixed in the token — no header needed |
The three credentials are not interchangeable. An MLflow token or a container token cannot call the REST API.
Inside a training container you create none of these yourself. The platform issues an MLflow token for each run, passes it in as an environment variable, and revokes it when the run ends.
The Bearer token of the DataOps integration is issued by the DataOps server, not by the platform. The platform stores it on the external source and sends it when it calls DataOps.