Team Templates

Team Templates#

Team templates define multi-agent configurations for collaborative tasks.

Location#

Team templates are stored in .pantheon/teams/:

.pantheon/
└── teams/
    ├── developer_team.md
    ├── research_team.md
    └── data_team.md

Template Format#

Basic Structure#

---
name: My Team
team_type: pantheon
agents:
  - name: agent1
    model: openai/gpt-4o
    instructions: First agent instructions.
  - name: agent2
    model: openai/gpt-4o-mini
    instructions: Second agent instructions.
---

# Team Description

This team works on collaborative tasks.

Frontmatter Fields#

Field	Required	Description
`name`	Yes	Display name for the team
`team_type`	No	Team type: `pantheon`, `swarm`, `sequential`, `moa`
`agents`	Yes	List of agent configurations
`model`	No	Default model for all agents
`orchestrator`	No	Orchestrator configuration (for pantheon teams)

Agent Configuration#

Each agent in the agents list:

agents:
  - name: researcher        # Required: agent name
    model: openai/gpt-4o    # Optional: override default model
    icon: 🔍                # Optional: display icon
    instructions: |         # Required: agent instructions
      You are a researcher.
      Search for information.
    toolsets:               # Optional: agent-specific tools
      - web_browse
    mcp_servers:            # Optional: MCP servers
      - github

Team Types#

Pantheon Team (Default)#

Orchestrator-managed team with dynamic task delegation:

---
name: Development Team
team_type: pantheon
orchestrator:
  model: openai/gpt-4o
  instructions: Coordinate the development process.
agents:
  - name: architect
    instructions: Design system architecture.
  - name: developer
    instructions: Implement features.
    toolsets:
      - file_manager
      - shell
  - name: reviewer
    instructions: Review code quality.
---

A team for software development projects.

Swarm Team#

Agents can hand off to each other dynamically:

---
name: Support Team
team_type: swarm
agents:
  - name: triage
    instructions: |
      Assess customer issues.
      Hand off to specialist or support.
    handoffs:
      - specialist
      - support
  - name: specialist
    instructions: Handle technical issues.
  - name: support
    instructions: Handle general inquiries.
---

Customer support with dynamic routing.

Sequential Team#

Agents process tasks in order:

---
name: Content Pipeline
team_type: sequential
agents:
  - name: researcher
    instructions: Research the topic.
  - name: writer
    instructions: Write the content.
  - name: editor
    instructions: Edit and polish.
---

Sequential content creation pipeline.

MoA Team (Mixture of Agents)#

Multiple agents contribute, then aggregate:

---
name: Analysis Team
team_type: moa
agents:
  - name: analyst1
    model: openai/gpt-4o
    instructions: Analyze from business perspective.
  - name: analyst2
    model: anthropic/claude-3-opus
    instructions: Analyze from technical perspective.
  - name: analyst3
    model: openai/gpt-4o
    instructions: Analyze from user perspective.
aggregator:
  model: openai/gpt-4o
  instructions: Synthesize all perspectives.
---

Multi-perspective analysis team.

Examples#

Developer Team#

---
name: Developer Team
team_type: pantheon
model: openai/gpt-4o
agents:
  - name: planner
    icon: 📋
    instructions: |
      You are a project planner.
      Break down tasks into steps.
      Create implementation plans.

  - name: developer
    icon: 👨‍💻
    instructions: |
      You are an expert developer.
      Write clean, tested code.
      Follow best practices.
    toolsets:
      - file_manager
      - shell
      - python_interpreter

  - name: reviewer
    icon: 🔍
    instructions: |
      You are a code reviewer.
      Check for bugs and issues.
      Suggest improvements.
    toolsets:
      - file_manager
---

# Developer Team

A team for software development with planning,
implementation, and review phases.

Research Team#

---
name: Research Team
team_type: sequential
agents:
  - name: searcher
    icon: 🔎
    instructions: Search for relevant information.
    toolsets:
      - web_browse

  - name: analyst
    icon: 📊
    instructions: Analyze and summarize findings.

  - name: writer
    icon: ✍️
    instructions: Write clear research reports.
    toolsets:
      - file_manager
---

# Research Team

Sequential research pipeline: search → analyze → report.

Data Analysis Team#

---
name: Data Team
team_type: pantheon
agents:
  - name: engineer
    icon: 🔧
    instructions: |
      Data engineering tasks.
      Clean and transform data.
    toolsets:
      - python_interpreter
      - file_manager

  - name: analyst
    icon: 📈
    instructions: |
      Statistical analysis.
      Generate insights.
    toolsets:
      - python_interpreter
      - notebook

  - name: visualizer
    icon: 🎨
    instructions: |
      Create visualizations.
      Design charts and dashboards.
    toolsets:
      - python_interpreter
      - notebook
---

# Data Analysis Team

Comprehensive data analysis with engineering,
analysis, and visualization capabilities.

Usage#

REPL:

pantheon cli --template developer_team

ChatRoom:

pantheon ui --auto-start-nats --auto-ui --template developer_team

Python API:

from pantheon.factory import load_team

team = load_team("developer_team")
response = await team.run("Build a REST API")

Best Practices#

Clear Roles: Each agent should have a distinct, well-defined role
Minimal Overlap: Avoid giving agents overlapping responsibilities
Appropriate Tools: Only give agents the tools they need
Right Team Type: Choose the team type that matches your workflow
Good Instructions: Provide clear, specific instructions for each agent