BaseAgent¶

The BaseAgent class is the foundation for all agents in AgentShip. It provides a complete framework for building AI agents with minimal boilerplate.

Overview¶

BaseAgent handles:

Agent initialization and configuration
Session management (automatic)
LLM integration via Google ADK
Response parsing and formatting
Tool support for extended functionality
Conversation context management

All agents in AgentShip inherit from BaseAgent, which provides the core functionality needed to interact with LLMs and manage conversations.

Class Definition¶

Key Methods¶

__init__(_caller_file, input_schema, output_schema): Initialize the agent with caller file path and schemas. Config is auto-loaded from the YAML file adjacent to the caller.
async chat(request: AgentChatRequest) -> AgentChatResponse: Main method for chatting with the agent. Handles session management and response formatting.
async run(user_id, session_id, input_data) -> BaseModel: Lower-level method for running the agent with schema validation.

Usage Example¶

from src.all_agents.base_agent import BaseAgent
from src.service.models.base_models import TextInput, TextOutput

# Create agent — config auto-loads from main_agent.yaml
class MyAgent(BaseAgent):
    def __init__(self):
        super().__init__(
            config_path=resolve_config_path(relative_to=__file__),
            input_schema=TextInput,
            output_schema=TextOutput
        )

# Use the agent
agent = MyAgent()
response = await agent.chat(request)

Session Management¶

Sessions are automatically managed by BaseAgent:

Sessions are created on first use
Conversation history is maintained
Supports both database and in-memory storage
Session IDs are scoped to user_id

Input/Output Schemas¶

Agents can define custom input/output schemas:

TextInput/TextOutput: Simple text-based I/O
Custom schemas: Define structured data for your use case
Validation: Pydantic validates all inputs automatically

Tool Support¶

Agents can extend functionality with tools:

Tools are automatically discovered
Tools are passed to the LLM
LLM decides when to use tools
Tool results are included in responses

See building-agents for more on creating agents with tools.