Core Components

Core Components#

Deep dive into Pantheon’s core building blocks.

Overview#

Pantheon is built on these core components:

  ┌─────────────────────────────────────────┐
  │              Your Application            │
  └─────────────────┬───────────────────────┘
                    │
  ┌─────────────────▼───────────────────────┐
  │     Agent / Team (Orchestration)         │
  └─────────────────┬───────────────────────┘
                    │
  ┌────────┬────────┼────────┬──────────────┐
  │        │        │        │              │
  ▼        ▼        ▼        ▼              ▼
Tools   Memory  Providers  Models       Config
  │        │        │        │              │
  └────────┴────────┴────────┴──────────────┘

Component Summary#

Agents

The fundamental unit of intelligence. An agent combines a model with instructions and tools.

Teams

Multiple agents working together. Supports various collaboration patterns (orchestrated, sequential, swarm).

Toolsets

Collections of tools that give agents capabilities like file access, code execution, and web search.

Memory

Conversation history and context management. Supports persistence and compression.

Providers

Abstractions for external tool sources (MCP servers, remote endpoints).

Relationships#

from pantheon.agent import Agent
from pantheon.team import Team
from pantheon.toolsets import FileManagerToolSet
from pantheon.memory import Memory
from pantheon.providers import MCPProvider

# Create an agent
agent = Agent(
    name="assistant",
    instructions="You are a helpful assistant.",
    memory=Memory()
)
# Add toolsets at runtime
await agent.toolset(FileManagerToolSet("files"))

# A team coordinates multiple agents
team = Team(
    name="dev_team",
    agents=[agent1, agent2, agent3]
)

# Providers connect to external tools via MCP
agent_with_mcp = Agent(
    name="github_agent",
    instructions="You help with GitHub operations."
)
await agent_with_mcp.mcp(
    name="github",
    provider=MCPProvider("npx -y @anthropic/mcp-server-github")
)

When to Use What#

Single Agent

Simple tasks, single domain expertise, straightforward workflows.

Team

Complex tasks, multiple expertise areas, tasks requiring review/iteration.

Custom Toolsets

When built-in tools don’t cover your needs.

MCP Providers

Integrating with external services, using community tools.