Adapter

Adapter Registry is a centralized library for fine-tuned adapters in Protean AI. Adapters represent the learned data on top of a base model, capturing task-specific knowledge, behavior, and optimization without duplicating the full model weights.

Adapters are first-class, deployable artifacts in Protean AI. The registry enables teams to manage fine-tuning projects, track lineage, compare runs, and safely publish the best adapter.

Adapter Registry provides the following advantages:

• Centralized catalog of all fine-tuned adapters
• Clear separation between base models and learned behavior
• Full lineage tracking across datasets, configurations, and runs
• Immutable release and deployment of adapters
• Built-in access control and enterprise governance

Adapter Configuration

To register an adapter, you define the adapter's identity and metadata.
An adapter is always associated with a base model and represents the outcome of one or more fine-tuning runs.

See the screenshot below for an example of adapter configuration.

Snapshot of Protean AI Platform

Name

The adapter name is a unique identifier.
It is used to reference the adapter during deployment. The name must be unique across all adapters and may contain alphanumeric characters.

Model

The Model field specifies the base model on which the adapter was trained. For external adapter, only models registered in the Model Registry can be selected, and the adapter can be deployed exclusively together with this base model. For the adapters published from fine-tuning runs, the model is automatically inferred from the fine-tuning configuration. This explicit linkage guarantees compatibility between the adapter weights and the underlying model architecture. The model field is read-only after creation.

Title

The title provides a short, human-readable identifier for the adapter. It is intended for quick recognition in catalog views and comparison workflows.

Description

The description provides a detailed explanation of the adapter's behavior and intended use. It should clearly state what the adapter was trained to do, describe the nature of the training data, outline the intended deployment scenarios, and highlight any known limitations or constraints. A clear description helps teams select the correct adapter and avoid misuse.

Lineage

Adapters in Protean AI maintain full lineage, on adapters that were produced by fine-tuning runs.

A lineage represents the complete evolution path of an adapter, including:

• Fine-tuning configuration
• Training dataset
• Dataset revision
• Training run and evaluation

Lineage is immutable and automatically maintained by the platform.

External vs Internal Adapters

Adapters can originate from two sources:

Internal Adapters

Internal adapters are produced by fine-tuning runs executed inside Protean AI. They are fully tracked with complete lineage, metrics, and dataset information, and are published directly from the evaluation steps of fine-tuning runs.

External Adapters

External adapters are imported via file upload using the HuggingFace LoRA format and are explicitly marked as external. This provides a way to deploy adapters that were not produced within Protean AI while keeping them manageable through the same registry.

Deployment

Adapters are first-class, deployable artifacts. An adapter can only be deployed by attaching it to its corresponding base model, and together they form a runnable inference unit. See Model Deployment Once deployed, the adapter follows the same runtime behavior, access controls, and operational rules as standard model deployments.

Only authorized users are permitted to deploy models and include adapters as part of a deployment.

Access Control

Access Control in Protean AI governs who can view, create, modify, and operate resources across the platform. It is designed for enterprise environments where security, isolation, and governance are mandatory.

Protean AI follows a principle of least privilege, ensuring users and systems are granted only the permissions required to perform their tasks.

Role→ Action↓	Admin	Model Admin	User	Owner	Viewer	Description
Create	Yes	Yes	Yes	NA	NA	Register or import adapters
Read	Yes	Yes	No	Yes	Yes	Read adapter metadata & deploy adapter for inferencing
Update	Yes	Yes	No	Yes	No	Edit adapter metadata
Delete	Yes	Yes	No	Yes	No	Deregister the adapter and remove it from the system
Manage Access	Yes	Yes	No	Yes	No	Grant or revoke permissions for users and groups.

Adapter Deletion

• Deleting an adapter from the registry also deletes all the artifacts from the storage.
• Adapter can only be deleted if it is not currently deployed.

Workflow

1. Select a base model from the Model Registry
2. Run fine-tuning to produce an adapter, and publish an adapter from an evaluation or import an external adapter
3. Deploy the adapter when ready

Result

Adapters are the bridge between raw models and production-ready intelligence. After registration, adapters become reusable, governed, and deployable assets.