Skip to main content

Understanding AI Agents

AI Agents in Airtrain are autonomous runtime entities that execute tasks, process decisions, and handle system interactions. This guide covers the core architecture and implementation patterns of Airtrain's agent system.

What is an AI Agent?​

An AI Agent is a software entity that:

  • Executes autonomous decision logic
  • Processes defined task workflows
  • Manages system and API interactions
  • Implements permission protocols
  • Controls resource allocation

Agent Components​

Each Airtrain agent implements three core components:

  1. Knowledge Base

    • Domain-specific knowledge
    • Learned patterns
    • Configuration settings

  2. Decision Engine

    • Task processing logic
    • Decision-making algorithms
    • Response generation

  3. Interaction Layer

    • User communication
    • System integration
    • API endpoints

Agent Types​

Airtrain implements two distinct agent classes: Task Agents and AI Employees. Task Agents handle atomic, single-purpose operations with strict boundaries, while AI Employees implement stateful interactions with complex memory management patterns. Understanding these implementation differences is crucial for optimal agent selection and deployment.

Task Agents​

Task Agents are specialized workers that execute atomic operations within defined boundaries. They implement a stateless execution model with predefined permission sets. When triggered, they execute their assigned task if permissions are satisfied, or return an error if access requirements aren't met.

  • Execution Model

    • Triggered for specific tasks
    • Can run in background or as cron jobs
    • Can be triggered by AI Employees
    • No persistent memory between tasks

  • Memory Scope

    • Limited to task lifetime
    • Starts fresh with each task
    • Must complete all persistence operations during task execution
    • Memory cleared after task completion

  • Permissions

    • Predefined, fixed permissions
    • Returns error if permission is missing
    • No ability to request additional permissions

AI Employees​

AI Employees implement a stateful agent pattern with persistent memory management and complex interaction handling. They maintain runtime context and can execute or delegate multiple task types. Unlike Task Agents, they implement interactive workflows and handle dynamic permission management.

  • Interaction Model

    • Triggered by user interface interactions
    • Can be assigned by users or other AI Employees
    • Maintains ongoing conversations and context
    • Can trigger Task Agents as needed

  • Memory Management

    • Maintains persistent memory
    • Automatically manages working vs persistent memory
    • Updates existing memories without explicit instructions
    • Retains context across multiple interactions

  • Permissions and Skills

    • Comes with predefined permissions and skill sets
    • Can request additional permissions from triggering user
    • Will reach back to initiator when lacking access
    • Can coordinate multiple tasks and resources

Implementation Differences​

Key architectural differences between Task Agents and AI Employees:

  1. Interaction Pattern

    • Task Agents: Stateless execution model
    • AI Employees: Stateful, interactive workflow

  2. Memory Handling

    • Task Agents: Runtime-scoped memory
    • AI Employees: Persistent state management

  3. Permission Management

    • Task Agents: Static permission validation
    • AI Employees: Dynamic permission requests

Next Steps​

Proceed to the implementation guide for detailed setup instructions, or explore our agent templates for quick deployment examples.