Key Concepts¶

LLMling and LLMling-Agent¶

LLMling-Agent extends LLMling's environment system to provide easy-to-configure agents:

graph TD
    A[LLMling-Agent] --> B[LLMling Core]
    A --> C[Language Models]
    B --> D[Tools & Resources]

LLMling Core Provides¶

Resource management (files, APIs, databases)
Tool registration and execution
Environment configuration
Template processing
Prompt management

LLMling-Agent Adds¶

Pure YAML-based agent configuration
Interactive chat sessions (CLI and web)
Model interaction handling
Session state management
History tracking

Core Strengths¶

YAML-First Philosophy¶

LLMling-Agent is built around the idea that AI agents should be: - Easily configurable through YAML - Fully reproducible - Version controllable - Human readable and verifiable - Shareable across teams and projects

This means you can define complete agents, their capabilities, and their environments in pure YAML without writing any code.

Universal Resource Access¶

Thanks to UPath integration, agents can work with resources from anywhere: - Local files - Remote URLs (http, https, s3, etc.) - Git repositories - Cloud storage - Any UPath-supported protocol

Dependencies are handled automatically - if an agent needs specific packages to interact with resources, they're installed on demand.

Usage Patterns¶

1. YAML-Based Configuration¶

Define complete agents in YAML without writing code:

# agents.yml
agents:
  web_assistant:
    model: openai:gpt-4
    environment:  # LLMling Config structure
      type: inline
      tools:
        open_browser:
          import_path: webbrowser.open
          description: "Open URL in default browser"
      resources:
        bookmarks:
          type: text
          content: |
            Python: https://python.org
            Docs: https://docs.python.org
    system_prompts:
      - "You are a helpful web assistant."

2. Interactive Sessions¶

Quick access to agents through CLI or web interface:

# CLI chat
llmling-agent chat web_assistant

# Web interface
llmling-agent launch

Both interfaces provide a consistent command system using slash commands:

/list-tools              # Show available tools
/enable-tool <name>      # Enable a tool
/disable-tool <name>     # Disable a tool
/show-resource <name>    # View resource content
/register-tool <path>    # Add new tool
/set-model <model>       # Change model
/history show           # View chat history
/stats                  # Show usage statistics
/help                   # Show all commands

3. Programmatic Usage¶

For integration into larger applications:

from llmling_agent import Agent

async with Agent.open_agent("agents.yml", "web_assistant") as agent:
    result = await agent.run("Open Python website")

Tools and Resources¶

LLMling-Agent leverages LLMling's resource system, providing agents with access to:

Tools¶

File operations
API interactions
System commands
Custom Python functions

Resources¶

File content
API endpoints
Database connections
Memory storage
Template systems

All tool and resource access is: - Configurable through YAML - Permission controlled - Usage tracked - Error handled

Agent Capabilities¶

Agents have role-based capabilities that control their access to: - Tool usage - Resource access - History viewing - Statistics access

Note: All agents are aligned with the Model Context Protocol (MCP), ensuring consistent interaction patterns across different models and configurations.

For detailed configuration options, see: - Agent Configuration - LLMling Environment Documentation