Virtualised deployment

In trac, model code is never copied into the platform. Models remain in their external repository and are fetched dynamically at runtime. The platform stores only a reference to the specific repository commit — making every deployed model version a permanent, recoverable asset without any additional archiving effort.

There are three steps to this process:

Import model	The platform scans but does not copy the model code. A metadata object is created containing the model schema and a reference to the repository commit. The model is now available for use on the platform.
Build flow (optional)	For multi-model calculations, models are assembled into a flow — a directed graph of model and data object keys. Schema consistency across connected models and datasets is validated before the flow is saved.
Run model / Run flow	To execute a job, the platform resolves the calculation contract: fetching model code from the repository at the referenced commit, reading input data from internal storage, and dispatching execution to the compute service. All assets are resolved from immutable references — the same job will produce the same result regardless of when it is run.

Note

This short video explains the virtualised deployment method in more detail.