Skip to content

Core Concepts

You're in Step 3 of the Learning Path

You've learned why LionAGI is different. Now let's understand the practical mechanics that make it work.

Understanding LionAGI's core abstractions is essential for building effective multi-agent workflows. These concepts work together to provide the parallel execution, memory isolation, and flexible orchestration that make LionAGI powerful.

Key Abstractions

Learning Order

Read these in order for best understanding:
1. Sessions & Branches (the foundation)
2. Operations (how work gets done)
3. Messages & Memory (how context works)

Sessions and Branches

  • Session: Workspace that coordinates multiple agents
  • Branch: Individual agent with memory and tools

Operations

  • Building blocks of workflows
  • Types: chat, communicate, operate, ReAct

Messages and Memory

  • How conversation state is managed
  • Memory isolation between branches

Tools and Functions

  • Extending agents with capabilities
  • Built-in and custom tools

Models and Providers

  • iModel abstraction for LLM providers
  • Supporting OpenAI, Anthropic, Ollama, etc.

The Mental Model

# Traditional: Agents talk to each other
agent1 → agent2 → agent3 → result

# LionAGI: Agents work in parallel, results synthesized
agent1 ↘
agent2 → synthesis → result
agent3 ↗

Architecture

Session (Workspace)
├── Branch (Agent 1)
│   ├── Messages (Memory)
│   ├── Tools
│   └── Model Config
├── Branch (Agent 2)
│   ├── Messages (Memory)
│   ├── Tools
│   └── Model Config
└── Graph (Workflow)
    ├── Operations
    └── Dependencies

Ready to Build Workflows?

Now that you understand the core concepts, it's time to see them in action:

Next: Patterns - Learn proven multi-agent workflow patterns
Or: Cookbook - Jump to complete working examples