An AI Agent Team With a Mission Control Board
How do you build an AI agent team with a mission control board? In this demo a made-up company, Acme, gets bug reports on GitHub and three agents handle them: a manager that classifies and supervises, a support agent that drafts the reply, and an engineering agent that reads the docs. After this lesson you can clone the demo repo, run it with a mock run or through OpenRouter, and define the boundaries each agent needs.
What this covers: the agent team, with the manager as orchestrator; the demo repository, its README and the board with its activity log; identity and authorization with Keycloak, a policy definition for each tool and approval, and traces with Langfuse; and bounding every agent by what it can do, what it cannot, and which tools it can reach, such as browser access or reading certain docs.
#aiagents #missioncontrol
Chapters
Transcript
The agent team
we have a company called ACME, and it is a made-up software company, you can say, and it gets bug report on GitHub. Okay, and it has two, three agents. One is a coordinator manager agent, and another one is the support agent, which will like the manager agent classify okay. This is a support ticket. the support agent can take care of it and reply for the customer, and then an engineering agent which will go and read your documentation and find out, and then it will answer or it will provide guidance in terms of what need to be done, and then the manager agent, which is more like the orchestrator, which will supervise and which will say, Okay, this piece of work is what you can do, and whether you have done it or not.
The demo repo
Okay, this example. What I am going to run right now, it is in GitHub repo. so you can check out as well as you can read this GitHub repo over here. and you can clone it and you can run and it has mock run as well. so for example, this is the read me. it has it has details like I'm using OpenRouter and you can you know run the demo as well as you can run the board in 8080 port, so I can quickly show how that board look like. So, this example I have created as I shown three AI agents, and I created the team for it, the manager agent and support resolver and the engineering analyst. And in the board, we have a couple of tasks which has been done. because I I already done the loop, and I will showcase this what these tasks are and how it runs in the board shortly, and it has some activity list on like what are all the tasks which has been done and what are all the steps which has been done so that anyone like human wanted to read the traces of what has been done, whether it is doing the right job, we will be able to do okay. Let's go back to the GitHub. as I said, it has details of how it is used, and this example uses some of the key things which is very important, like how you can verify who is doing what action, whether this person is eligible to do that, that you can do with Keycloak, and you know it has policy definition of what what each tool does and what approval processes, and then it has traces with Langfuse. So you can check it out and you can run it locally if you want, and so on with that.
Bound every agent
You can extend the mission control for your use cases. So if you wanted to build your own mission control, these are some of the key areas you need to you need to build, like what it can do and what is their boundary, like what it can do, what it can't do that you need to define clearly and what it can access in terms of tools, whether it can do browser access or whether it can just read certain documentation that you need to clearly mention, then it is bounded to running only the things it is it is defined.
This lesson comes with Pro, $149 a year: Get Pro. This lesson is cut from the free Lightning session Build a Mission Control for AI Agents. The Live Labs build the same system with you, free, on Thursdays: see the Live Labs.