Agent Commons
Overview
Integrate the possibility to visually define agents with this module. Combined with other assets like GenAI Commons, Conversational UI and the connectors for models of OpenAI, Amazon Bedrock and Mendix Cloud GenAI, you have all the things you need to build smart, agentic AI apps with Mendix.
This module contains a domain model, pages and toolbox operations to define and call agents. It allows developers and prompt engineers to collaborate and build agentic systems.
See the Agent Builder Starter App for a full app with a template support agent for an example of how to use this module.
Create agents grounded in private knowledge like static reference guides and that can take action with tools, to call APIs or to retrieve live data in the app, like tickets.
- Use any of the supported connectors to leverage a variety of large language models in your agent from OpenAI, Amazon Bedrock, Mendix Cloud or Bring Your Own.
- Take advantage of all the capabilities of Conversational UI, Prompt Management and the GenAI Commons modules. Agents created with Agent Commons are fully integrated & compatible with these components.
Explore the full potential of generative and agentic AI in your Mendix applications with this starter app template and start building today!
Contact Us
For support and questions, feel free to reach out via email genai-components-feedback@mendix.com or Community Slack.
Documentation
This module contains a domain model, pages and toolbox operations to define and call agents. It allows developers and prompt engineers to collaborate and build agentic systems.
View documentation on Mendix docs.
General Setup
- Download the latest versions of the following modules from the Marketplace and follow the setup instructions: GenAICommons, ConversationalUI. A connector that is compatible with GenAI Commons is required for the interactions with your LLMs of choice.
- In the project security of your app, add the module role 'AgentCommons.AgentAdmin' to the user roles that are intended to define and refine Agents.
- Add the 'Agent_Overview' page (USE_ME > Agent Builder) to your Navigation or add the 'Snippet_Agent_Overview' to a page that is already part of your Navigation. Now the Agents can be defined at runtime.
Define Agents
- Run the app (locally or in the cloud).
- Create and edit agents at runtime. Write the prompts, use variable placeholders, add microflows as tools and connect knowledge bases to define the tasks and cabapilities of the agents.
- Configure the deployed model (LLM) that the agent should use, and connect the agent to it.
- Create various versions of the agents and test in an isolated setup to rapidly iterate, compare and evaluate the agentic behavior
- Set a version as "in use" to allow it to be called from the actual logic in the app.
Call an Agent from App Logic
Use toolbox operations "Call Agent" and "Create Chat for Agent" to use an agent in a microflow and make it part of your app logic. Take a look at our documentation or download the Agent Builder Starter App from the Marketplace for a detailed example of how to use the "Call Agent" activity in an action microflow of a chat interface.
Releases
* Small UI fix for rejected tool calls in test chat interface
* We changed the order in which messages are added to the request to prevent possible issues where the order of messages passed to the model would have been incorrect.
This is a replacement of the earlier released module 3.0.0, which has been unpublished.
The module has been aligned with the latest changes in ConsumedKnowledgeBase and ConsumedMCPService. This is a breaking change and might need migration to prevent loss of data and functionality. For migration steps, please refer to the migration guide here: https://docs.mendix.com/appstore/modules/genai/genai-for-mx/migration-guide/
User Access and Approval is a new GenAI Commons and ConversationalUI feature to make tool calls visible to users and provide control for users to decide if tools should be executed or rejected. The agent editor makes this configurable at runtime for tools (including MCP tools). For knowledge bases it can be decided whether they are visible to the user or not. Per default all tools and knowledge base retrievals are hidden for users (same behavior as in previous versions). You can optionally provide a display title and description to be shown to the user.
When applying User Access and Approval, the call agent operations behave differently than in previous versions, because the tool call response is sent back to the microflow without the final assistant response:
- For chat-agents that are used in the ConversationalUI module inside of action microflows nothing needs to change, only if pre- or postprocessing was added that should not happen multiple times per sending the user message
- For task-agents, you'd need to implement your own UI and logic to show the tools on your pages and make the confirmation/rejection of tools possible. There is an example in the GenAI Showcase app as well as in the how-to create your first agent on Mendix Docs.
This module version is compatible with the GenAI Commons 6.0.0, ConversationalUI 6.0.0 and MCPClient 3.0.1 versions.
Moved duplicated styling classes into ConversationalUI and renamed classes if they were used by Atlas classes to prevent overwriting. Ensure to update ConversationalUI as well.
* Protocol version 2025-03-26 can now be selected when registering connections to MCP servers to configure Streamable HTTP transport method.
* In the Agent Overview page agents created in runtime are now marked with an 'In App' label.
* A different mechanism to show available context entities has been implemented to minimize warnings because of Exceeded Request State Size Thresholds.
* When selecting an MCP server to consume tools from inside of the Agent Details, a link to the MCP server configurations overview page can be used as a shortcut to specify a new configuration.
* A bug was fixed, where test cases in Draft would not show values entered in draft mode if opened another time.
* The module was upgraded to 10.24.13.
Changes made to support new capability in ConversationalUI v5.3.0: Observability for AI Traces
Static image widget instances were replaced with Image widgets.
Support for Mistral Deployed Models has been added to the Agent Builder.
Synchronization of MCP tools when opening an agent has been added to task queue to be completed in the background instead. Current synchronization status is now displayed for each MCP tool.