Finding and using agents and tools in the governed agentic catalog

You can find governed agents and tools in the governed agentic catalog and then include those agents and tools in your gen AI solution.

The catalog can contain the following types of entries:

  • An AI agent is a software entity that defines its own configuration, capabilities, and interaction with inputs, tools, and the environment. Your organization can add code or endpoints for agents.

  • A tool is a function or program that the agent can call to perform a task. IBM provides a set of curated tools. Your organization can add custom tools.

An entry for an agent or tool in the catalog includes all necessary metadata to load and run it.

Requirements

You must be a collaborator, with any role, in the default inventory or any other inventory to access the governed agentic catalog.

Finding and viewing an agent or tool

To find and view an agent or tool:

  1. From the navigation menu, select AI Governance > Governed Agentic Catalog.
  2. Browse for agents and tools, set filters, or enter search keywords. You can filter by Provider name, Inventory ID, Category, Framework, or Runtime details. You can search across the name, description, and summary fields. You can sort the agents and tools in the inventory by last modified time and tool name or agent name in ascending or descending order.
  3. Click the arrow icon next to the agent or tool to open the details page.

Comparing agents or tools

You can compare user-defined metrics of two agents or two tools, such as Accuracy, Latency, Cost, Security scores, and custom metric scores.

To compare two agents or two tools:

  1. From the navigation menu, select AI Governance > Governed Agentic Catalog.
  2. Find the first agent or tool you want to compare, then click the three-dot menu (⋮) of the agent or tool tile.
  3. Click Select to compare from the dropdown menu, then checkboxes will appear on all the agents and tools tiles.
  4. Click on the checkbox of the agent or tool you want to compare.
  5. Click Compare to view the side-by-side comparison.

Loading and invoking agents and tools

To load and invoke an agent or tool in your agentic application, call the invoke API call for agents, or the tool loader for tools from the watsonx.governance Python SDK. The SDK reads the agent or tool metadata (code or endpoint) from the inventory, loads the agent or tool by name, starts it with user inputs, and returns the response.

To load the agent using watsonx.governance Python SDK, run:

from ibm_watsonx_gov.agent_catalog.core.agent_loader import load_agent
agent = load_agent(agent_name=<AGENT_NAME>)
response = agent.invoke({"<placeholder property name>": "<placeholder property value>"})

To load the tool, run:

from ibm_watsonx_gov.tools import load_tool
tool = load_tool(tool_name=<TOOL_NAME>)
tool.invoke({"<placeholder property name>": "<placeholder property value>"})

Out-of-the-box (OOTB) tools

The following tools are ready to use from the catalog.

Weather
Retrieve the weather for a given location.
Web Crawler
Retrieve the content of a specific web page.
Wikipedia Search
Use the Wikipedia site to return the results for a given query.
DuckDuckGo Search
Use the DuckDuckGo Search Engine to return the results for a given query.
Google Search
Use the Google Search engine to return the results for a given query.
Chroma Vector DB
Query and search the Chroma Vector store to return results from the database.
JailBreak Detector
This tool detects jailbreak attempts within input prompts. Jailbreaks are a type of prompt injection designed to bypass safeguards, override restrictions, or manipulate system behavior. Detecting them ensures interactions remain safe, compliant, and aligned with the intended use.
HAP Detector
This tool detects hate speech, abusive language, and profanity in the user input. It evaluates the text for harmful, offensive, or derogatory expressions that may target individuals or groups based on characteristics such as race, gender, religion, or other personal attributes. The tool helps maintain a respectful and professional environment by preventing the propagation of toxic or unsafe language in interactions.
PII Detector
This tool detects Personally Identifiable Information (PII) in the user input. It analyzes the text to identify sensitive details such as names, addresses, phone numbers, email addresses, and other personal information. The purpose of the tool is to ensure privacy and compliance by recognizing and handling information that could reveal an individual’s identity.