BHANUJBHANUJ
v1.0.3

Create Managed Assets in Studio

Use Studio to create and manage the versioned prompts, models, datasets, and evaluator configurations that AI Governance Platform owns. This is distinct from reporting observed runtime assets from an external AI system.

Overview

Open Assets in Studio, then choose the catalog for the asset you want to manage. Use this workflow when the platform should own the governed record; use Observed Asset Ingestion when your application remains the source of runtime truth and only reports the evidence it used.

Creation is versioned

Creating a prompt, model, or dataset records an immutable managed version. Change the content or runtime configuration by creating a new version—do not overwrite historical governance evidence.

Create Prompt

  1. Open Assets → Prompt Catalog.
  2. Select Create Prompt.
  3. Enter the prompt name, version, template, and any template variables.
  4. Select Create Prompt to save the managed draft version.

The prompt template and its declared variables are part of that immutable version. When the prompt changes, open the managed prompt record and create a new version instead of changing the original.

Register Model

Before registering a provider-hosted model, configure an enabled Runtime Connection for that provider in Settings → Runtime Connections. Studio uses the connection to discover provider model IDs while keeping credentials private.

  1. Open Assets → Model Catalog and select Register Model.
  2. Select the runtime provider and enter the governed model name.
  3. Select an active catalog connection and the provider model ID it discovers.
  4. Enter a version and context window, then set any supported runtime defaults.
  5. Select Register Model to save the managed draft version.

Studio exposes only verified runtime controls for the selected provider and model. Registering the model records the exact configuration that governed work can later reference; it does not deploy or serve that model.

Register Dataset

  1. Open Assets → Datasets and select Register Dataset.
  2. Enter a name, version, description, and schema version.
  3. Upload a UTF-8 CSV, JSONL, or NDJSON file.
  4. Select Upload and register.

Studio calculates the checksum and record count, stores the immutable bytes in the configured object store, and registers the resulting draft version. From the dataset detail page, freeze a version for evaluation or activate it for governed use.

Configure Evaluator

Evaluation providers are shipped adapters; Studio configures tenant-scoped installations rather than uploading provider code.

  1. Open Assets → Evaluation Providers.
  2. Select New Installation and choose a provider type.
  3. Set the display name, scope, and adapter configuration.
  4. Provide secret references, never raw credentials.
  5. Select Validate Connection, then Create Installation.

After Creation

Open the asset record to inspect its versions, references, lineage, and audit history. Create a new version when the governed input changes, then use the appropriate lifecycle action before selecting it in governed work.

Next, follow the Experiment Management tutorial to combine managed inputs in an evaluation, or read Asset Catalogs for the ownership and provenance model.