Skip to main content

​
Agents

Agents are AI systems packaged as Docker containers that can execute tasks. The Plato agent framework provides:
  • Typed Configuration: Pydantic-based config with automatic secret loading
  • Docker Packaging: Run agents in isolated containers
  • Trajectory Capture: ATIF-format logging of agent actions
  • Registry: Register and discover agents by name

​
Installation

uv add plato-sdk-v2

​
Quick Example

from plato.agents import BaseAgent, AgentConfig, Secret, register_agent
from typing import Annotated

class MyAgentConfig(AgentConfig):
    model_name: str = "anthropic/claude-sonnet-4"
    api_key: Annotated[str, Secret(description="API key")]

@register_agent("my-agent")
class MyAgent(BaseAgent[MyAgentConfig]):
    name = "my-agent"
    description = "My custom agent"

    async def run(self, instruction: str) -> None:
        # Access typed configuration
        model = self.config.model_name
        api_key = self.config.api_key

        # Execute task logic
        ...

        # Write trajectory
        await self.write_trajectory(trajectory)

​
Core Concepts

​
BaseAgent

The base class for all agents. Provides:
  • Typed config access via generics
  • Logger instance
  • Trajectory writing
  • Version detection

​
AgentConfig

Configuration base class that:
  • Extends pydantic-settings for env var loading
  • Marks secret fields with Secret annotation
  • Provides JSON schema generation

​
Secret Annotation

Mark sensitive fields that should be loaded from environment: 
api_key: Annotated[str, Secret(description="API key")]
# Loaded from ANTHROPIC_API_KEY env var

​
Registration

Register agents for discovery: 
@register_agent("my-agent")
class MyAgent(BaseAgent[MyAgentConfig]):
    ...

# Retrieve agents
from plato.agents import get_agent, get_registered_agents
agent_cls = get_agent("my-agent")

​
What’s Next?

Creating Agents

Build custom agents with typed configuration

Running Agents

Execute agents in Docker containers