Warehouse Agent AI Hub demo

This project includes a basic demo of the AI hub, as shown during the Developer Community webinar in August 2026.

The concept is simple, there is an agent class with tools to interact with our InterSystems IRIS instance, including:

• Retrieving product information with SQL Queries
• Logging stock loss with an ObjectScript tool
• Ordering new stock via an Interop production with an objectscript tool that creates a business service.

Setup

To use AI Hub, you currently (as of Aug 2026) need to download a separate IRIS edition from https://evaluation.intersystems.com/Eval/early-access/AIHub. At time of writing, the IRIS community editions available are version 126. You need to download the docker.tar.gz edition as shown in the image below.

If there is a more recent iris-community version available, you will need to update the image defined in Dockerfile to reflect this. M-chip mac users will also need to select the ARM version and update the docker file.

after downloading you will need to load the file into your docker engine:

docker load -i /path/to/iris-community-2026.3.0AI.126.0-docker.tar.gz

You will also need an LLM provider API key to run this. I have been using AWS Bedrock, however, its easy to change the models in the agent class (Information on other providers is in the AI hub docs).

The docker-compose project currently reads .env, so you need to create this file to start up the project. Put your API key with in here with an environmental variable name (e.g. in .env.example), and then edit the agent class to use this provider and environmental variable.

After loading the docker file, you can then start the container as normal with

docker-compose up -d 

Once started the demo will automatically populate. The only other thing to do is start the production (Warehouse.Interop.Production)

Usage

Agent

To use the agent class, use the following:

set agent = ##class(Warehouse.AI.Agent).%New()
set sc = agent.%Init() 
w sc 
set session = agent.CreateSession()
set response = agent.Chat(session, "What tools do you have>")
do ##class(%AI.System).RenderMarkdown(response.Content)

Or start a basic chat loop with:

do agent.ChatLoop(session)

Suggested prompts:

- What products need reordering? 
- Which products are selling the best? 
- Can you order 20 memory foam pillow pairs for me
- We had an accident in the warehouse and broke 5 dinner sets, can you log this

MCP Server

The same tools available to the agent are also available as an MCP server. To start this MCP server, you need to create a web-application, this can be done from System Administration -> Security -> Applications -> MCP Servers, as shown in the webinar. I’ve also provided Warehouse.AI.MCPSetup.cls to show the code way of doing this:

do ##class(Warehouse.AI.MCPSetup).Run()

You then need to start the iris-mcp-server binary from a bash terminal:

iris-mcp-server -c config.toml run

This configuration file is mounted in the container. Depending on the agent you are using to connect to the MCP server, the exact connection method will vary. However, the project-scoped .mcp.json provides the connection for Claude Code as well as some other agents. For other agents, the connection definition will look very similar, although the location of the definition will vary.