liman-core 0.1.0a1

Liman AgentOps Core Library

Liman Core

Core library of the Liman framework providing low-level building blocks for declarative YAML-based agent workflows with custom DSL.

Features

• Declarative YAML Configuration: Define agents using simple YAML manifests
• Multi-language Support: Built-in localization for prompts and descriptions
• Node-based Architecture: Compose workflows from LLM, Tool, and custom nodes
• Edge DSL: Connect nodes with conditional expressions and function references

Node Types

LLMNode

Wraps LLM requests with system prompts and tool integration.

kind: LLMNode
name: assistant
prompts:
  system:
    en: |
      You are a helpful assistant.
    es: |
      Eres un asistente útil.
tools:
  - calculator
nodes:
  - target: analyzer
    when: "result == 'success'"
  - target: error_handler
    when: "result != 'success' and retry_count < 3"

ToolNode

Defines function calls for LLM tool integration.

kind: ToolNode
name: calculator
description:
  en: |
    Performs mathematical calculations
func: my_module.calculate
arguments:
  - name: expression
    type: str
    description:
      en: Mathematical expression to evaluate

Installation

pip install liman_core

Quick Start

from liman_core import LLMNode, ToolNode

# Load from YAML
llm_node = LLMNode.from_yaml_path("agent.yaml")

# Create from dict
tool_spec = {
    "kind": "ToolNode",
    "name": "calculator",
    "description": {"en": "Math tool"},
    "func": "math.sqrt"
}
tool_node = ToolNode.from_dict(tool_spec)

# Compile and invoke
llm_node.compile()
result = llm_node.invoke("What is the square root of 16?")

Edge DSL

Connect nodes using the custom DSL for conditional execution:

nodes:
  # Simple target reference
  - analyzer

  # Conditional execution with variables
  - target: success_handler
    when: "status == 'complete' and errors == 0"

  # Logical operators
  - target: retry_handler
    when: "failed and (retry_count < 3 or critical == true)"

  # Function reference for complex logic
  - target: custom_validator
    when: "validators.check_output"

Supported DSL features:

• Constants: true, false, numbers, strings
• Comparisons: ==, !=, >, <
• Logical operators: and/&&, or/||, not/!
• Function references: module.function for custom logic
• Parentheses: For expression grouping

API Reference

All nodes inherit from BaseNode and provide:

• .from_dict() - Create from dictionary
• .from_yaml_path() - Load from YAML file
• .compile() - Prepare node for execution
• .invoke() / .ainvoke() - Synchronous/asynchronous execution
• .print_spec() - Display formatted specification