Key Features

What you can do

Agents as Core Primitives

LLMs configured with instructions, tools, and guardrails to perform tasks autonomously.

Handoffs Between Agents

Agents can delegate specific tasks to specialized sub-agents, enabling complex multi-agent workflows.

Function Tools

Any Python function can be wrapped as a tool with automatic schema generation from function signatures and docstrings.

Built-in Tracing

Integrated tracking to visualize, debug, and evaluate agentic flows and workflows.

Agent Loop

Automatic handling of tool invocation, result processing, and continuation until task completion.

Guardrails

Configurable validation of agent inputs and outputs to ensure correctness and safety.