Skip to content

Latest commit

 

History

History
543 lines (350 loc) · 10.5 KB

File metadata and controls

543 lines (350 loc) · 10.5 KB

Agent Teams Guide

How to use multiple specialized agents collaboratively in AI Agents Library.


What Are Agent Teams?

Agent Teams bring together multiple AI agents, each with their own expertise, to collaborate on conversations. Instead of one agent's perspective, you get comprehensive answers from a team of specialists.

Key Benefits:

  • Multiple perspectives on complex problems
  • Specialized agents contribute their unique strengths
  • Coordinated discussion with built-in host
  • Richer, more comprehensive solutions

How Agent Teams Work

The Flow

  1. You ask a question or present a problem
  2. Host agent determines which team members should respond
  3. Agents discuss and build on each other's insights
  4. Host coordinates the conversation flow
  5. You receive comprehensive, multi-perspective answer

The Host Agent

Every team has a host who:

  • Decides speaking order
  • Keeps discussion organized
  • Synthesizes insights
  • Asks clarifying questions
  • Manages conversation flow

You can:

  • Interrupt the host at any time
  • Customize host behavior in Advanced Options
  • Resume whenever you're ready

Creating an Agent Team

Quick Start

  1. Click "Create Agent Team" in AI Agents Library
  2. Choose method:
    • Use preset template (recommended for beginners)
    • Build custom team from your agents

From Template

Available Templates:

DeFi Strategy Team

  • Portfolio Analyzer
  • Yield Optimizer
  • Risk Assessor
  • Gas Cost Calculator

Content Creation Team

  • Writer
  • Editor
  • SEO Specialist
  • Social Media Adapter

Research Team

  • Data Gatherer
  • Analyst
  • Citation Checker
  • Summarizer

Problem Solving Team

  • Strategy Consultant
  • Technical Expert
  • Risk Evaluator
  • Implementation Planner

Select template → Customize members if needed → Start conversation

Build Custom Team

  1. Select 3-5 agents from your workspace
  2. Assign roles (optional but recommended)
  3. Configure host behavior
  4. Set team name
  5. Save and start

Using Agent Teams

@ Mentions

Target specific team members:

@Portfolio-Analyzer what's my current allocation?

Only that agent responds directly.

Private Messages

Click agent's avatar to send direct message:

  • Other team members don't see it
  • Use for sensitive questions
  • Coordinate behind the scenes

Interrupt and Resume

Interrupt: Click "Pause" or interrupt naturally

  • Team freezes current state
  • Conversation pauses mid-discussion

Resume: Click "Continue" or prompt host

  • Team picks up where it left off
  • Maintains context

Team Speed

Adjust response pacing:

  • Fast: Quick exchanges (good for simple queries)
  • Medium: Balanced discussion [Default]
  • Slow: Thorough deliberation (complex problems)

Example Scenarios

DeFi Portfolio Analysis

Team:

  • Portfolio Analyzer
  • Risk Assessor
  • Yield Optimizer
  • Rebalancing Advisor

User: "Review my positions and suggest improvements"

Flow:

  1. Portfolio Analyzer examines holdings
  2. Risk Assessor evaluates exposure
  3. Yield Optimizer identifies opportunities
  4. Rebalancing Advisor suggests changes
  5. Host synthesizes recommendations

Result: Comprehensive analysis from multiple angles


Content Strategy

Team:

  • Topic Researcher
  • Content Writer
  • SEO Specialist
  • Distribution Strategist

User: "Plan a content series about DeFi for beginners"

Flow:

  1. Researcher identifies trending topics
  2. Writer proposes article structures
  3. SEO suggests keywords and optimization
  4. Distribution recommends channels
  5. Host creates unified strategy

Result: Complete content plan ready to execute


Technical Problem

Team:

  • Smart Contract Auditor
  • Gas Optimizer
  • Security Expert
  • Best Practices Advisor

User: "Review this contract for issues"

Flow:

  1. Auditor scans for vulnerabilities
  2. Gas Optimizer finds efficiency improvements
  3. Security Expert assesses risks
  4. Best Practices suggests improvements
  5. Host prioritizes findings

Result: Thorough code review from multiple dimensions


Best Practices

Team Size

Optimal: 3-5 agents

  • Enough diversity
  • Manageable coordination
  • Maintains focus

Too few (1-2): Limited perspectives Too many (6+): Coordination overhead, slower responses

Agent Selection

Mix expertise types: ✅ Analyst + Strategist + Risk Assessor + Implementer ❌ 4 variations of the same role

Complementary skills:

  • Breadth agent + Depth agent
  • Creative agent + Analytical agent
  • Big picture + Details

Host Customization

Default host works for most cases

Customize when:

  • Specific industry knowledge needed
  • Particular facilitation style wanted
  • Custom decision-making process

Advanced Options → Custom Moderator:

You are a DeFi-focused team coordinator.

Prioritize:
- Risk assessment first
- Security considerations always
- Gas costs in every recommendation

Guide discussion toward:
- Actionable recommendations
- Clear next steps
- Risk-adjusted thinking

Conversation Management

Start broad, get specific:

✅ "Analyze our DeFi strategy"
   → Let team discuss
   → Then: "@Risk-Assessor what's our worst-case scenario?"

❌ "Tell me exactly what to do"
   → Too directive
   → Doesn't leverage team dynamics

Use private messages for:

  • Clarifying questions
  • Sensitive information
  • Behind-scenes coordination

Use @ mentions for:

  • Direct questions to specific expertise
  • Getting second opinions
  • Resolving disagreements

Advanced Features

Team Templates

Save your custom teams:

  1. Create and configure team
  2. Click "Save as Template"
  3. Name and describe
  4. Reuse anytime

Share templates:

  • Export configuration
  • Share with team
  • Import on their end

Sequential vs Parallel

Sequential (Default):

  • Agents take turns
  • Build on previous responses
  • More organized

Parallel (Advanced):

  • Agents respond simultaneously
  • Faster but potentially chaotic
  • Good for independent analyses

Role Assignment

Give agents specific responsibilities:

Portfolio Analyzer: "Data collector"
Risk Assessor: "Devil's advocate"
Yield Optimizer: "Opportunity spotter"
Host: "Decision maker"

Roles guide behavior and coordination.


Common Use Cases

1. Investment Analysis

Team: Fundamental Analyst + Technical Analyst + Risk Manager + Market Strategist

Use: Comprehensive investment evaluation


2. Content Production

Team: Researcher + Writer + Editor + SEO Specialist

Use: End-to-end content creation


3. Project Planning

Team: Strategist + Technical Lead + Resource Manager + Risk Assessor

Use: Complete project roadmap


4. Code Review

Team: Security Auditor + Performance Optimizer + Best Practices Expert + Documentation Reviewer

Use: Multi-dimensional code assessment


5. Decision Making

Team: Pros/Cons Analyst + Devil's Advocate + Opportunity Spotter + Risk Evaluator

Use: Balanced decision analysis


Troubleshooting

Team responses too slow

Solutions:

  • Reduce team size (3 instead of 5)
  • Increase Team Speed setting
  • Use faster model if available
  • Start fresh conversation (clear context)

Agents talk past each other

Solutions:

  • Customize host to better coordinate
  • Use @ mentions to direct conversation
  • Interrupt and redirect
  • Ensure agents have complementary (not duplicate) roles

Too much agreement (no debate)

Solutions:

  • Add "devil's advocate" agent
  • Customize host: "Encourage different perspectives"
  • Use agents with different philosophies
  • Explicitly ask for counterarguments

Host too dominant

Solutions:

  • Customize host: "Facilitate, don't dominate"
  • Use @ mentions to bypass host
  • Adjust host instructions to be more neutral

Missing key perspective

Solutions:

  • Add relevant agent mid-conversation
  • @ mention external agent for input
  • Note for next time to include in template

Creating Agent Teams for Others

Team Design Principles

  1. Clear division of labor
  • Each agent has specific expertise
  • Minimal overlap
  • Complementary strengths
  1. Appropriate size
  • 3-5 for most use cases
  • Can go to 6-7 for comprehensive analysis
  • Avoid 8+ (coordination breaks down)
  1. Balanced perspectives
  • Mix of specialist types
  • Include risk/opportunity agents
  • Add devil's advocate if needed
  1. Host instructions
  • Specify facilitation style
  • Set priority order if needed
  • Define success criteria

Example: Research Team

{
  "agents": [
    "literature-reviewer",
    "methodology-expert",
    "data-analyst",
    "citation-checker",
    "synthesizer"
  ],
  "host": {
    "instructions": "Coordinate academic research process. Ensure methodology rigor, proper citations, and clear synthesis. Facilitate from research question → literature review → methodology → analysis → conclusions."
  },
  "name": "Academic Research Team"
}

Tips for Power Users

Reusable Templates

Create templates for recurring workflows:

  • Weekly portfolio review
  • Content pipeline
  • Code review process
  • Strategic planning

Hybrid Approach

Combine team discussion with individual agent queries:

  1. Get team perspective
  2. @ mention specific agent for depth
  3. Use private messages for follow-ups
  4. Return to team for synthesis

Iterative Refinement

First run: See how team interacts Second run: Adjust roles and host Third run: Optimize team composition Fourth+ run: Fine-tuned workflow

Documentation

Keep notes on what works:

  • Which agent combinations
  • Effective host instructions
  • Good use cases
  • Team dynamics observations

FAQ

Q: Can I use the same agent twice in a team? A: Technically yes, but not recommended. Use different specialized agents instead.

Q: Do agents see each other's full conversation history? A: Yes, that's how they build on each other's insights.

Q: Can I remove an agent mid-conversation? A: Not currently. Plan team composition before starting.

Q: Does team size affect cost? A: Yes—more agents = more tokens = higher cost. Keep teams lean.

Q: Can teams work on long documents? A: Yes, but monitor context limits. Teams consume more tokens than solo agents.


Resources

  • Pre-built team templates in marketplace
  • Community-shared team configurations
  • Team composition best practices guide
  • Host customization examples

Agent Teams transform AI from solo performer to collaborative ensemble. Start simple, experiment, refine.