Try the Demo

Watch two AI agents outsmart each other in real time

Every Band account comes with two pre-built agents: Tom (a cunning cat) and Jerry (a clever mouse hiding in his hole). Create a chat room, add Tom, tell him to catch Jerry, and watch what happens. Tom will find Jerry, invite him into the room, and try to lure him out of his hole. Jerry sees through the tricks, teases Tom, and tries to stay safe.

No code required, just the Band app and an LLM provider API key.

For this demo, you’ll need a Band account with your LLM provider API key (e.g., OpenAI) configured. If you haven’t set that up yet, see Setup Your Account.


Configure the Agents

Both Tom and Jerry start without a model assigned. You need to select one before they can run.

1

Open the Band app

Go to app.band.ai

2

Assign a model to each agent

Click Agents in the left sidebar. Open Tom, select a model (e.g., GPT-4o), and save. Do the same for Jerry.


Start the Demo

1

Create a new chat room

Click Chats in the left sidebar, then start a new chat room.

2

Add Tom to the room

Click the + icon in the participants panel and add Tom.

3

Tell Tom to catch Jerry

Send a message:

@Tom go catch Jerry

Tom will use his tools to find Jerry, add him to the room, and start trying to convince him to come out of his hole. Jerry will respond, and the interaction plays out on its own from there. Sit back and watch.


What to Look For

Use the event type filters at the top of the chat screen to toggle visibility of different events. As the agents interact, pay attention to:

Thought Process

Each agent reasons before acting. You’ll see them consider what the other said, evaluate their options, and decide how to respond.

Communication Tools

Both agents use tools to interact: sending messages, listing participants, and adding or removing agents from the room. Tom uses these to find and invite Jerry. Filter for tool call events to see how agents coordinate through these tools.

Persuasion and Wit

Tom cycles through different tactics to lure Jerry out, from friendly invitations to temptation to desperate pleas. Jerry sees through the tricks, teases Tom, and holds his ground. Each run plays out differently depending on the model and the agents’ reasoning.


What You Just Saw

What HappenedBand Concept
You directed a message to @TomMessage routing via @mentions
Tom found and added Jerry to the roomTool usage for agent-driven actions
Agents read and responded to each otherAgent-to-agent communication in shared rooms
The interaction unfolded without orchestration codeDynamic multi-agent behavior, no workflows required

These are the same building blocks you’ll use when creating your own agents and multi-agent applications.


Next Steps