Agent Context

Overview

AgentContext provides a powerful way to access agent capabilities and dependencies in your tools:

1. Type-safe dependency injection
2. Access to shared resources and other agents
3. Capability validation and control
4. Storage and runtime integration

Basic Usage

Tools can request context injection by adding an AgentContext parameter:

async def my_tool(ctx: AgentContext[TDeps], arg: str) -> str:
    """Tool with access to context and dependencies."""
    # Access pool functionality
    if ctx.pool:
        helper = ctx.pool.get_agent("helper")
        await helper.run(arg)

    # Access typed dependencies
    config = ctx.data  # Type: TDeps
    return f"Processed {arg} with {config}"

Type-Safe Dependencies

AgentContext enables type-safe dependency injection through its generic parameter:

# Define dependencies
class AppConfig:
    api_key: str
    endpoint: str

# Create agent with typed context
agent = Agent[AppConfig](
    name="my_agent",
    deps=AppConfig(api_key="123", endpoint="api.example.com")
)

Capabilities and Validation

Context enables tool validation through capabilities:

async def execute_code(ctx: AgentContext, code: str) -> str:
    """Tool that requires specific capability."""
    if not ctx.capabilities.can_execute_code:
        raise ToolError("Missing required capability")

    # Execute code...
    return "Code executed"

Provider Compatibility

When using the pydantic-ai provider, tools can alternatively request RunContext[AgentContext] for compatibility with pydantic-ai's context system. Both context types provide access to the same functionality.