Using integrations¶

Integrations in OPAQUE are preconfigured, reusable components that can be shared across workspaces and workflows. They let teams securely connect to external systems (like LLMs or data sources) without having to re-enter credentials or config details every time.

Think of integrations as ready-to-use building blocks: define them once, and make them available for anyone in your workspace to use in workflows.

Supported integration types¶

OPAQUE currently supports three types of integrations:

• Data connectors (retrievers): Connect to external data sources to retrieve structured or unstructured data at runtime.

  Example: An Azure AI Search Retriever integration that queries an enterprise search index.

• LLMs (language models): Connect to large language models for text processing, analysis, or generation.

  Example: An OpenAI Service integration or a vLLM Service integration.

• MCP API Tool: Configure an external tool that can be invoked by an Agent node running in tool mode using the Model Context Protocol (MCP).

Create a new integration¶

To create a new integration:

1. Go to the Integrations page in the OPAQUE and click New Integration.

   

2. Choose the integration type—either a Data connector or an LLM—and click Next.

   • Data connector: Create when you want agents to pull information from an external data source (currently Azure AI Search).
   • LLM: Create when you want agents to use a large language model (OpenAI or vLLM).

   

3. Select the type of integration to add for use in agentic workflows.

   If you chose Data connector ... If you chose LLM ...

   . . . the available options are Azure AI Search Retriever, PostgeSQL Connector, and MCP API Tool.

   

   Note

   Although the MCP API Tool is currently grouped under data connectors, it is conceptually different. It provides a generic interface for agents to interact with external APIs, rather than connecting to a specific data source. This grouping is temporary and will be corrected in a future release.

   ... you can configure Anthropic, OpenAI Service, or vLLM services.

   

4. Configure the agent. Give it a clear name and description so teammates can quickly recognize its purpose. Each integration type includes fields specific to its role.

   Name integrations clearly

   Choose names and descriptions that make sense to others in your organization. For example, use “Customer Search Index” instead of “Azure Test 1.” Clear naming helps teammates recognize and reuse the right integration without confusion.

   Configure Azure AI Search RetrieverConfigure OpenAI or vLLM

   When configuring Azure AI Search Retriever, you can supply the following fields:

   • API key: Usually required to connect securely.
   • API version: Optional; defaults to the latest supported version.
   • Search index name: Typically required to query your data.
   • Search service name: Optional; identifies the Azure Search service hosting your index.

   

   When configuring an LLM agent, you’ll want to specify:

   • Model name
   • Temperature
   • Maximum number of tokens
   • API key/URL (depending on the service)
   • Context prompt

   Any fields you pre-specify here will be locked and uneditable for downstream users.

   

5. Click Save. The new integration appears on the Integrations page.

   

Share an integration with workspaces¶

Once created, an integration can be shared with one or more workspaces so it’s available in their workflows. Follow these steps:

1. On the Integrations page, select the integration from the list and click Share with workspaces. You’ll need to be the integration’s creator or a workspace admin to share it.

   

2. Choose one or more workspaces and click Share integration.

   

You’ll see a confirmation notification that your new integration has been shared with the selected workspace(s).

To confirm, go to the target workspace and open the Integrations tab. You should see the shared integration listed.

Click the more () icon, to open a drawer that shows the agent’s configuration details.

Once shared, integrations appear in the Nodes panel of your workflow, with a icon next to them. From there, they behave just like nodes you configure yourself — you can connect them in workflows, define prompts, and add guardrails as needed.