CoolAIRegistry Community Edition

Enter your e-mail address to download the installation files (.zip)

What is it?

The primary function of the Cool AI Registry Portal is to serve as a central repository for AI agents, track their versions, and automatically check their system prompts against established AI policies (Guidelines). It helps ensure that agents, whether developed in-house on ODC or externally, adhere to predefined compliance rules.

The challenge

As we add more agents into our organization, the challenge to oversee and govern them all becomes greater. So, the question is: how can we keep our agents in check?
Organizations need a solution to register and manage their AI agents. An extra challenge is that organizations will probably not use just OutSystems Agent Workbench agents. It is not unlikely that other platforms are being used, such as Microsoft Azure Foundry. This is why CoolAIRegistry also supports Copilot agents and agents from other platforms.

Hybrid Architecture

CoolProfs, in fact, proposes a hybrid architecture in which we choose to centrally host the LLM models in Azure Foundry. This location is typically best managed by technical teams. They take care of monitoring, security, model selection and potential fine‑tuning, as well as legislation and policies. In addition, LLM costs are easy to manage here.

In a Business layer, developers with domain knowledge are best positioned to define the agent’s tasks – think system prompts, grounding data, and any user prompts. You can build agents with the OutSystems Agent Workbench, but Azure/Copilot‑based agents are also possible.

Key Concepts

CoolAIRegistry distinguishes between:

• ODC Agents: Agents developed and managed within the OutSystems Development Cloud (ODC). Their data is synchronized from the ODC environment.
• External Agents: Agents registered manually, with their details and system prompts provided directly in the portal.

AI Policies (Guidelines): A set of rules and instructions that AI agents must follow. The portal allows for the creation and management of these guidelines. The system can analyze guideline documents (PDFs) to automatically generate specific compliance checks.

Compliance Check: A process, either automated via AI or performed manually, that evaluates an agent’s system prompt against the active AI Policy. The results determine if an agent is “Compliant” or “Not Compliant”.

Human Review: A workflow for compliance reviewers to manually inspect, approve, or reject an agent’s compliance status, with the ability to override automated checks.