Skip to content

Overview

Kagura AI Agents are the core building blocks of the Kagura AI system. These modular components enable developers to create flexible, reusable, and scalable workflows for various applications.

What is a Kagura AI Agent?

A Kagura agent is a self-contained unit designed to:

  • Perform specific tasks in an AI workflow.
  • Manage state data through YAML-defined models.
  • Interact with other agents via state sharing and orchestration.

Agents are configurable and extensible, making them adaptable to diverse use cases.

Core Agent Properties

  1. Independence: Each agent operates as a standalone module with its own state and logic.
  2. Collaboration: Agents can share states and collaborate within workflows.
  3. Extensibility: Agents can include custom tools and pre/post-processing hooks.
  4. Type Safety: All data is validated using Pydantic models, ensuring consistency.

Agent Types

Kagura offers three primary agent types, each designed for specific use cases:

Atomic Agent:

  • Core stateful agent for LLM tasks.
  • Supports pre- and post-processing hooks for flexibility.
  • Generates structured outputs with type-safe validation.

Tool Agent:

  • Focuses on non-LLM data processing.
  • Ideal for data transformations, API integrations, and fast execution.

Workflow Agent:

  • Manages workflows involving multiple agents.
  • Enables conditional routing, error recovery, and progress monitoring.
  • Uses predefined Atomic Agent and Tool Agent to orchestrate tasks.

For a detailed description of each agent type, see the Agent Types document.

Agent Configuration

Agents are configured using YAML files, which define their roles, state models, and interactions. The basic structure of an agent is as follows:

agents/
└── agent_name/
    ├── agent.yml         # Core configuration
    ├── state_model.yml   # State definition
    └── tools.py          # Custom tools

To learn more about setting up agents, refer to the Agent Configuration document.

Workflow Integration

Agents are orchestrated to form complex workflows. The workflow agent handles state sharing, data transformations, and conditional logic between agents. For more information on orchestrating agents, see the Workflow Agent section.

Custom Tools

Agents can be extended with custom tools for specialized tasks. These tools integrate seamlessly into the agent’s lifecycle, providing additional functionality. See the Custom Tool section for more details.

Next Steps

To start building with Kagura agents:

These guides will help you effectively utilize Kagura agents in your projects.