Running Agents

This guide covers different ways to interact with LLMling agents.

Command Line Interface

The llmling-agent command provides several subcommands:

# List available commands
llmling-agent --help

# Manage agent configurations
llmling-agent add my-agents agents.yml   # Add configuration
llmling-agent list                       # List available agents
llmling-agent set my-agents             # Set active configuration

# Run agents
llmling-agent run my-agent "What's the system status?"  # Single query
llmling-agent chat my-agent                             # Interactive chat
llmling-agent run my-agent -p check_system              # Use predefined prompt

# Multiple agents
llmling-agent run "agent1,agent2" "Analyze this"        # Run multiple agents

# View history
llmling-agent history show                              # Show recent conversations
llmling-agent history stats --group-by model            # Show usage statistics

Advanced CLI options:

# Override model
llmling-agent run my-agent --model openai:gpt-4 "Complex query"

# Output formatting
llmling-agent run my-agent --output-format json "Query"

# Environment selection
llmling-agent run my-agent --environment prod.yml "Query"

# Debug mode
llmling-agent chat my-agent --debug

Interactive Chat Sessions

Chat sessions provide an interactive interface with enhanced features:

$ llmling-agent chat my-agent
Started chat with my-agent
Available tools: 5 (5 enabled)

You: What tools can you use?
Assistant: I can use these tools:
- read_file: Read file contents
- analyze_code: Run code analysis
- get_stats: Get system statistics

Web Interface

Launch the web interface for a graphical experience:

# Start web interface
llmling-agent launch

# Custom server settings
llmling-agent launch --host 0.0.0.0 --port 8000

# Create shareable link
llmling-agent launch --share

The web interface provides: - File upload for configurations - Agent selection and management - Real-time chat with streaming responses - Tool state visualization - Conversation history - Debug logging

Programmatic Usage

Use agents in your Python code:

from llmling_agent import Agent

# Basic usage
async with Agent.open_agent("agents.yml", "my-agent") as agent:

    # Basic query
    result = await agent.run("What's the system status?")
    print(result.data)

    # Streaming responses
    async with await agent.run_stream("Long analysis...") as stream:
        async for chunk in stream:
            print(chunk.content, end="", flush=True)

    # Managing tools
    agent.tools.enable_tool("search")  # Enable specific tool
    agent.tools.disable_tool("edit")   # Disable specific tool
    print(agent.tools.list_tools())    # See all tools and their states

    # Working with conversation history
    history = agent.conversation.get_history()  # Get current history
    agent.conversation.clear()  # Clear history

    # Add context for next message
    await agent.conversation.add_context_from_file("data.txt")
    await agent.conversation.add_context_from_resource("api_docs")

For more details, see: - CLI Reference - Web Interface Guide - API Documentation