Concepts

Mental models for building real agents with Liman

Liman runs agents as a graph of nodes. Each node does one thing. The framework connects them, runs them, and remembers where it left off. To build confidently, you need two mental models:

Execution Model

How nodes run: the three layers, graph traversal, and sequential vs parallel execution.

State

What gets remembered: multi-turn conversations, persistence, and crash recovery.

Last updated on

3. OpenAPI Integration

Automatically generate tools from OpenAPI specifications

Execution Model

How Liman builds and executes the AI agent - the four layers, graph traversal, and execution flow