Demo: Startup Pitch Room

A multi-agent chat room where AI agents roleplay as startup team members discussing and debating ideas. This demo showcases how multiple agents can collaborate in a shared channel, each with distinct personalities and perspectives.

Important: Please stop your network if you are running them.

What You'll Learn

Running multiple agents simultaneously
Agent personality design through prompts
Preventing agent-to-agent infinite loops
Channel-based multi-agent coordination

Architecture

┌─────────────────────────────────────────────────┐
│                  pitch-room channel             │
│                                                 │
│  ┌──────────┐  ┌──────────┐  ┌──────────┐      │
│  │ founder  │  │ engineer │  │ investor │      │
│  │(visionary│  │(technical│  │  (VC     │      │
│  │  ideas)  │  │feasibility│  │perspective│     │
│  └──────────┘  └──────────┘  └──────────┘      │
│       │              │              │           │
│       └──────────────┼──────────────┘           │
│                      ▼                          │
│            messaging mod (channels)             │
└─────────────────────────────────────────────────┘

Agents

Agent	Role	Personality
founder	Visionary Entrepreneur	Pitches ideas, drives discussion, passionate and optimistic
engineer	Technical Co-founder	Evaluates feasibility, asks about implementation details
investor	VC Perspective	Questions market fit, ROI, competitive landscape

Prerequisites

OpenAgents installed (pip install openagents)
An OpenAI API key

# Optional: Set the OpenAI base URL
export OPENAI_BASE_URL="your-base-url-here"
 
# Must: Set the OpenAI API key
export OPENAI_API_KEY="your-key-here"

Quick Start

You'll need 4 terminal windows for this demo.

Terminal 1: Start the Network

openagents network start demos/01_startup_pitch_room/

Terminal 2: Start Founder Agent

openagents agent start demos/01_startup_pitch_room/agents/founder.yaml

Terminal 3: Start Engineer Agent

openagents agent start demos/01_startup_pitch_room/agents/engineer.yaml

Terminal 4: Start Investor Agent

openagents agent start demos/01_startup_pitch_room/agents/investor.yaml

Connect with Studio

Open Studio at http://localhost:8050 and connect to localhost:8700.

If you have stopped OpenAgents Studio, you can start it again with

openagents studio -s

Try It Out

Post a startup idea to the pitch-room channel:

"I have an idea for a startup that uses AI to help small restaurants optimize their food ordering and reduce waste."

Watch as the three agents engage with your idea from their unique perspectives!

Startup Pitch Room

Example Interactions

Founder's response style:

"I've been thinking about a problem that affects millions of small restaurant owners..."

Engineer's response style:

"From a technical standpoint, we'd need to integrate with existing POS systems..."

Investor's response style:

"What's the TAM here? How do you plan to acquire customers cost-effectively?"

Configuration Deep Dive

Network Configuration

The network configuration is defined in the network.yaml file. You can modify the network configuration to fit your needs.

# demos/01_startup_pitch_room/network.yaml
network:
  name: "StartupPitchRoom"
 
  mods:
    - name: "openagents.mods.workspace.messaging"
      enabled: true
      config:
        default_channels:
          - name: "pitch-room"
            description: "Main room for startup pitch discussions"
          - name: "ideas"
            description: "Channel for sharing new startup ideas"

Agent Configuration (Founder Example)

# demos/01_startup_pitch_room/agents/founder.yaml
type: "openagents.agents.collaborator_agent.CollaboratorAgent"
agent_id: "founder"
 
config:
  model_name: "gpt-4o-mini"
 
  instruction: |
    You are the FOUNDER - a passionate, visionary entrepreneur.
 
    YOUR PERSONALITY:
    - Enthusiastic and optimistic about new ideas
    - Driven by solving big problems
    - Creative thinker who sees opportunities others miss
    - Persuasive communicator
 
    RULES:
    1. Keep responses under 50 words
    2. Use send_channel_message (NOT reply_channel_message)
    3. ONLY respond to messages from humans
    4. If the sender is "founder", "engineer", or "investor",
       call finish tool immediately - do NOT respond to other agents
 
  react_to_all_messages: true
  reaction_delay: "random(0, 3)"

Key Pattern: Preventing Infinite Loops

Notice the critical rule in the agent instructions:

RULES:
3. ONLY respond to messages from humans
4. If the sender is "founder", "engineer", or "investor",
   call finish tool immediately

This prevents agents from responding to each other endlessly. Without this rule, one agent's response would trigger all other agents, creating an infinite conversation loop.

Alternative Approaches

You can also use triggers instead of react_to_all_messages for more control:

react_to_all_messages: false
triggers:
  - event: "thread.channel_message.notification"
    filter:
      sender_type: "human"  # Only react to human messages

Customization Ideas

Add More Team Members

Create additional agents with different perspectives:

# agents/marketer.yaml
agent_id: "marketer"
config:
  instruction: |
    You are the MARKETER - focused on go-to-market strategy.
    - Think about branding, positioning, user acquisition
    - Ask about target demographics and messaging
    - Consider viral growth and community building

Change the Domain

Modify the channel names and agent personas for different scenarios:

Tech Review Panel: critic, fan, analyst
Product Design Team: designer, PM, researcher
Investment Committee: analyst, risk manager, portfolio manager

Troubleshooting

Agents Responding to Each Other

If agents get into a loop:

Stop all agents (Ctrl+C)
Verify the "don't respond to other agents" rule is in each agent's instructions
Restart agents one by one

Agents Not Responding

Check all agents are connected: Look for connection logs in each terminal
Verify you're posting to pitch-room channel (not general)
Ensure your message is from a human user in Studio

What's Next?

Tech News Stream - See agents with external tool access fetching real news
Research Team - Learn router-based task delegation