Agents

Agents in Protean AI (also referred to as Assistants) are configurable, reusable AI execution units that combine a deployed model, instructions, knowledge sources, and optional tools into a single operational artifact.

Agents allow teams to move beyond raw model inference and define repeatable AI behavior for applications, workflows, and users. Instead of re-prompting and re-configuring models for every interaction, an Agent encapsulates intent, constraints, and execution context in one place.

Agents are first-class, deployable resources in Protean AI and can be shared, versioned, and governed with enterprise-grade access control.

Agents provide the following advantages:

• Encapsulate model behavior, system instructions, and configuration in a single artifact
• Enable repeatable and consistent AI behavior across applications and users
• Integrate models, datasets, and tools without application-level wiring
• Support controlled publishing, sharing, and lifecycle management
• Reduce prompt duplication and operational complexity

Agent Configuration

An Agent definition describes how the AI behaves and operates during interactions. It specifies which model to use, how it should respond, what knowledge it can access, and which tools it may invoke. An Agent definition includes the following components.

Name

The Agent name is a human-readable identifier used throughout the platform. It must be unique within its scope and should clearly reflect the Agent's purpose or role.

Examples:

• Customer Support Assistant
• Kubernetes Troubleshooting Agent
• Internal Documentation Helper

Instructions

Instructions define the system prompt for the Agent. This is the primary mechanism for controlling behavior, tone, boundaries, and decision logic.

Instructions typically include:

• The Agent's role and responsibilities
• Behavioral constraints
• Output formatting rules
• Domain-specific guidance

Well-defined instructions ensure predictable and safe behavior across all interactions.

Model

Each Agent is bound to a deployed model. The selected model determines the Agent's fundamental capabilities, such as reasoning quality, tool usage, or response style.

Only compatible deployed models can be selected. If a model does not support certain features (such as tool calling), the platform automatically restricts incompatible configuration options.

Creativity (Temperature)

Creativity controls the randomness of the model's output.

• Lower values produce more deterministic, precise responses Useful for code generation, analysis, and structured outputs.
• Higher values increase variability and expressiveness Useful for brainstorming or creative writing.

Creativity can only be adjusted after the Agent is initially saved to ensure configuration stability.

MCP Servers (Tools)

Agents can be connected to MCP (Model Context Protocol) servers, which enable tool calling and external actions.

MCP servers may be:

• Internal (provided by Protean AI, such as SQL access)
• External (registered by the user or organization)

Tool availability is automatically validated against the selected model's capabilities. If a model does not support tool calling, MCP servers are disabled to prevent misconfiguration.

Knowledge (Datasets)

Agents can reference one or more datasets as knowledge sources. These datasets are used during inference to ground responses in domain-specific or proprietary information.

Knowledge datasets can be added or removed at any time without redefining the Agent itself, enabling iterative refinement without redeployment.

Common use cases include:

• Internal documentation assistants
• Policy or compliance agents
• Product or customer knowledge bots

Agent Lifecycle

Agents follow a clear and controlled lifecycle to support safe iteration and enterprise governance.

Draft

Newly created Agents start in a Draft state. Draft Agents can be edited freely but cannot be shared or used by others.

Publish

Publishing an Agent creates an immutable, deployable version. Only published Agents can be shared and used across teams.

Publishing locks:

• Instructions
• Model selection
• Tool configuration

Subsequent changes require creating a new revision.

Revision

Editing a published Agent creates a new revision. This is in draft status. This ensures historical versions remain intact and auditable while allowing controlled evolution.

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 / Create an agent
Read	Yes	No	No	Yes	Yes	View agent and use it in chat
Update	Yes	No	No	Yes	No	Update agent metadata
Delete	Yes	No	No	Yes	No	Remove agent
Manage Access	Yes	No	No	Yes	No	Grant or revoke permissions for users and groups.

Workflow

1. Create a new Agent.
2. Define instructions and select a deployed model.
3. Configure creativity and optional MCP servers.
4. Add knowledge datasets if required.
5. Save and test the Agent.
6. Publish the Agent for use and sharing.

Result

After publishing, the Agent becomes available for interactive use, integration with applications, and controlled sharing across teams. Agents provide a stable abstraction layer that allows applications to consume AI behavior without managing prompts, tools, or models directly.