Introduction

Welcome to agent-air, an open-source Agent SDK and Runtime built in Rust. agent-air provides everything you need to build, deploy, and run intelligent agents across multiple environments: terminal applications with rich TUI interfaces, desktop applications, and server-side deployments.

What is agent-air?

agent-air is a comprehensive framework that abstracts the complexities of building LLM-powered agents. At its core, it provides:

SDK Components

  • Agent Architecture - A flexible foundation for building agents with customizable behaviors, state management, and event-driven execution
  • LLM Provider Abstraction - Unified interface supporting multiple providers (Anthropic, OpenAI) with seamless switching between models
  • Tool System - Define and register tools that your agent can invoke, with full async support and type-safe schemas
  • Context Management - Intelligent conversation handling with automatic context window management and compaction strategies
  • Extension System - Extend your agent with custom widgets, commands, and tools through a modular plugin architecture

Runtime Capabilities

  • Terminal Runtime - Full-featured TUI (Text User Interface) with 45+ built-in themes, customizable layouts, and interactive widgets
  • Desktop Runtime - Deploy agents as standalone desktop applications
  • Server Runtime - Run agents as backend services for API-driven interactions

Key Features

  • Multi-Provider Support - Switch between Anthropic Claude and OpenAI models without code changes
  • Streaming Responses - Real-time token streaming for responsive user experiences
  • Session Management - Create, switch, and persist conversation sessions
  • Token Tracking - Monitor and optimize token usage across conversations
  • Rich Theming - Over 45 built-in themes with full customization support
  • Keyboard Bindings - Configurable key bindings with modal editing support
  • Slash Commands - Built-in and custom slash commands for quick actions

Architecture Overview

agent-air follows a layered architecture:

  1. Core Layer - Agent lifecycle, state management, and event loop
  2. LLM Layer - Provider abstraction, session management, and context handling
  3. Tool Layer - Tool registration, execution, and result handling
  4. UI Layer - Widgets, themes, layouts, and rendering pipeline
  5. Extension Layer - Plugin system for custom functionality

Who is this for?

agent-air is designed for developers who want to:

  • Build intelligent agents with conversational AI capabilities
  • Create interactive assistants and AI-powered tools
  • Develop LLM-powered applications with rich user interfaces
  • Deploy agents across terminal, desktop, or server environments
  • Integrate multiple LLM providers with a unified API

Use Cases

  • Development Assistants - Build coding assistants that help with development tasks
  • Data Analysis Tools - Create agents that can query, analyze, and visualize data
  • Customer Support - Deploy conversational agents for automated support
  • Workflow Automation - Build agents that orchestrate complex workflows
  • Research Tools - Create assistants for research and information gathering

Next Steps