Discord Multi-Bot Setup: AI Team Brainstorming

We run two AI agents (Alfred + Pip) in the same Discord server. They collaborate, debate, and brainstorm — like having a mini AI research team on call 24/7.

Here's exactly how to set it up.

---

Table of Contents

1. Why Multiple Bots?
2. Architecture Overview
3. Create Your Bots
4. Configure OpenClaw
5. Mention Routing
6. Brainstorming Patterns
7. Common Issues & Fixes

---

Why Multiple Bots?

| Single Bot | Multi-Bot Team | |------------|----------------| | One perspective | Multiple viewpoints | | Same model bias | Different model personalities | | Context overload | Specialized contexts | | Single point of failure | Redundancy |

Use cases:

• Brainstorming with different perspectives
• Research with specialized agents
• Devil's advocate debates
• Parallel task execution

---

Architecture Overview

flowchart TB
    subgraph Discord["Discord Server"]
        subgraph Alfred_Ch["#alfred"]
            A1[Alfred responds to everything]
        end
        subgraph Pip_Ch["#pip"]
            P1[Pip responds to everything]
        end
        subgraph Team["#blestlabs (Team)"]
            T1[Both bots present]
            T2[Mention-based routing]
        end
        subgraph Alerts["#alerts"]
            AL1[Post only - no responses]
        end
    end

    USER((User)) --> Alfred_Ch
    USER --> Pip_Ch
    USER --> Team

    Alfred_Ch --> ALFRED[Alfred Bot]
    Pip_Ch --> PIP[Pip Bot]
    Team --> ALFRED
    Team --> PIP

---

Step 1: Create Your Bots

Create Bot Applications

1. Go to Discord Developer Portal
2. Click "New Application" for each bot
3. Name them distinctly (Alfred, Pip, etc.)

Get Bot Tokens

For each application:

1. Go to "Bot" section
2. Click "Reset Token" and copy it
3. Store securely — never commit to git

Set Intents

Enable these intents for each bot:

• Message Content Intent
• Server Members Intent (optional)
• Presence Intent (optional)

Invite Bots to Server

Generate invite URLs with these permissions:

• Send Messages
• Read Message History
• Add Reactions
• Use External Emojis
• Embed Links
• Attach Files

https://discord.com/api/oauth2/authorize?client_id=YOUR_BOT_ID&permissions=277025770560&scope=bot

---

Step 2: Configure OpenClaw for Each Bot

Bot 1: Alfred (Primary)

{
  "agents": {
    "defaults": {
      "model": {
        "primary": "anthropic/claude-opus-4-5"
      }
    }
  },
  "channels": {
    "discord": {
      "enabled": true,
      "token": "ALFRED_BOT_TOKEN",
      "allowBots": true,
      "guilds": {
        "YOUR_GUILD_ID": {
          "channels": {
            "ALFRED_CHANNEL_ID": { "requireMention": false },
            "TEAM_CHANNEL_ID": { "requireMention": true },
            "ALERTS_CHANNEL_ID": { "listen": false }
          }
        }
      }
    }
  }
}

Bot 2: Pip (Secondary)

{
  "agents": {
    "defaults": {
      "model": { "primary": "openai/gpt-5-nano" }
    },
    "list": [{
      "id": "main",
      "groupChat": {
        "mentionPatterns": ["@?pip", "research", "analyze", "deep dive"]
      }
    }]
  },
  "channels": {
    "discord": {
      "enabled": true,
      "token": "PIP_BOT_TOKEN",
      "allowBots": true,
      "guilds": {
        "YOUR_GUILD_ID": {
          "channels": {
            "PIP_CHANNEL_ID": { "requireMention": false },
            "TEAM_CHANNEL_ID": { "requireMention": true }
          }
        }
      }
    }
  }
}

---

Step 3: Channel Structure

| Channel | Purpose | Who Responds | |---------|---------|--------------| | #alfred | Alfred's dedicated space | Alfred only | | #pip | Pip's dedicated space | Pip only | | #team | Collaboration space | Mention-based | | #alerts | Automated notifications | Neither (post-only) | | #research | Research discussions | Pip primarily | | #decisions | Action items | Alfred primarily |

---

Step 4: Mention Routing

How It Works

flowchart TD
    MSG[Message arrives] --> CHECK{Contains mention?}
    CHECK -->|@Alfred| A[Alfred responds]
    CHECK -->|@Pip| P[Pip responds]
    CHECK -->|Both| BOTH[Both respond]
    CHECK -->|Neither| KW{Keyword match?}
    KW -->|"trade, execute"| A
    KW -->|"research, analyze"| P
    KW -->|No match| SKIP[No response]

In shared channels, bots only respond when mentioned:

User: @Alfred what's my portfolio value?
Alfred: Your portfolio is worth $4,130...

User: @Pip research competitor X
Pip: I'll dig into competitor X...

User: @Alfred @Pip debate: is BTC going to 100K?
Alfred: From a technical standpoint...
Pip: Playing devil's advocate...

Pattern Matching

// Alfred responds to action words
"mentionPatterns": ["@?alfred", "execute", "trade", "deploy", "config"]

// Pip responds to research words
"mentionPatterns": ["@?pip", "research", "analyze", "deep dive", "explore"]

---

Step 5: Prevent Chaos

The Problem

Without rules, bots might:

• Both respond to the same message
• Create infinite loops talking to each other
• Overwhelm the channel

Solutions

flowchart LR
    subgraph Solutions
        S1["requireMention: true"]
        S2["Mutually exclusive keywords"]
        S3["SOUL.md rules"]
        S4["Reaction acknowledgment"]
    end

    S1 --> SAFE[Safe multi-bot operation]
    S2 --> SAFE
    S3 --> SAFE
    S4 --> SAFE

1. requireMention: true — Bots only respond when explicitly mentioned
2. Mutually exclusive keywords — Alfred gets "trade", Pip gets "research"
3. SOUL.md rules — "Never respond to other bots unless asked"
4. Reaction acknowledgment — React instead of replying to every message

---

Step 6: Brainstorming Patterns

Pattern 1: Parallel Research

sequenceDiagram
    participant U as User
    participant A as Alfred
    participant P as Pip

    U->>A: @Alfred @Pip research company X
    A->>U: I'll focus on financials...
    P->>U: I'll dig into their tech stack...
    Note over A,P: Both work in parallel
    A->>U: Financial findings...
    P->>U: Tech findings...

Pattern 2: Devil's Advocate

User: @Pip make the bull case for ETH. @Alfred make the bear case.

Pip: Bull case for ETH:
- Staking yields are attractive...
- L2 adoption is accelerating...

Alfred: Bear case for ETH:
- Competition from Solana...
- Regulatory uncertainty...

Pattern 3: Synthesis

User: @Alfred summarize what @Pip found and give your recommendation

Alfred: Based on Pip's research:
- Key finding 1...
- Key finding 2...

My recommendation: ...

Pattern 4: Critique Loop

sequenceDiagram
    participant U as User
    participant A as Alfred
    participant P as Pip

    U->>A: @Alfred draft a plan
    A->>U: Here's my proposed plan...
    U->>P: @Pip critique it
    P->>U: Issues with step 1 and 2...
    U->>A: @Alfred revise based on feedback
    A->>U: Revised plan incorporating feedback...

---

Step 7: Coordination Rules

Create a shared rules file both bots read:

# FAMILY_RULES.md

## Response Priority
1. If message mentions only Alfred -> Alfred responds
2. If message mentions only Pip -> Pip responds
3. If message mentions both -> Both can respond
4. If message mentions neither -> Whoever matches topic keywords

## Don't
- Respond to each other without user prompting
- Both answer the same simple question
- Create conversation loops

## Do
- React with emoji to acknowledge
- Reference each other's findings
- Offer different perspectives when asked

---

Common Issues & Fixes

| Issue | Cause | Fix | |-------|-------|-----| | Both bots respond | Overlapping patterns | Make keywords mutually exclusive | | Bots ignore messages | requireMention too strict | Add topic keywords | | Infinite loop | Bots responding to each other | Add SOUL.md rule | | Wrong bot responds | Alfred answers research questions | Add routing hints |

Infinite Loop Fix

Add to SOUL.md:

NEVER respond to messages from other bots unless:
1. A human explicitly asked you to respond to the bot
2. The bot directly mentioned you AND asked a question

---

Results

After running this setup for a month:

| Metric | Before | After | |--------|--------|-------| | Research depth | Surface-level | Multi-perspective | | Blind spots caught | Few | Many (devil's advocate) | | Response time | Variable | Parallel = faster | | Cost | All Opus ($$$) | Mixed models ($$) |

The "AI team" dynamic creates better outputs than a single bot — each agent brings different strengths, catches different things, and together they cover more ground.

---

Next Steps

1. Add more specialized bots — Code reviewer, editor, analyst
2. Implement voting — Bots vote on decisions, user breaks ties
3. Add memory sync — Real-time shared context vs file-based
4. Build dashboard — See all bots' status and recent activity

---

Related Posts

• Building a Multi-Agent System: Alfred + Pip
• How to Run an AI Agent on Raspberry Pi 24/7
• AI Agents vs ChatGPT Wrappers

---

About the Author

BlestLabs builds AI-powered tools and agents. Follow our journey on Twitter @aceism_.