by Gimle Labs
Create an Agent Use the Creator Use the Monitor

Use the Agent Creator

The Agent Creator is an interactive wizard that generates a complete agent for you. Instead of manually creating files, you describe what you want and let AI design the tools and configuration.

What is the Agent Creator?

The creator is a meta-agent that:

  1. Takes your description of what you want the agent to do
  2. Designs appropriate tools for that purpose
  3. Generates all the necessary files (config, task, template, tools)
  4. Creates a ready-to-run agent directory

It uses sonnet-latest by default for generation, as this model excels at code generation tasks.

Step 1: Start the Wizard

hugin create

The wizard will guide you through several prompts.

Step 2: Enter Agent Name

? Agent name: weather_assistant

Requirements:

Step 3: Describe What Your Agent Should Do

? What should this agent do?
> Help users check weather forecasts. It should be able to look up current
> weather and forecasts for any city, and provide helpful advice about
> what to wear or whether to bring an umbrella.

Tips for good descriptions:

Vague Better
"Help with weather" "Look up current weather and 5-day forecasts for any city, provide clothing recommendations"
"Do data stuff" "Read CSV files, calculate statistics, generate summary reports with charts"
"Research agent" "Search the web for information, summarize findings, cite sources"

Step 4: Choose the LLM Model

? LLM model for the agent: haiku-latest

The model your created agent will use (not the creator itself).

Hugin supports models from three providers:

Provider Examples Requires
Anthropic haiku-latest, sonnet-latest, opus-latest ANTHROPIC_API_KEY
OpenAI gpt-4o, gpt-4o-mini OPENAI_API_KEY
Ollama qwen3:8b, llama3.2-latest, llama3.3-70b Local Ollama installation

The default is haiku-latest which is fast and works well for most tasks.

See all available models: Check src/gimle/hugin/llm/models/model_registry.py for the complete list of pre-configured models.

Adding custom models: To add a new model, edit model_registry.py and register it with the appropriate provider class (AnthropicModel, OpenAIModel, or OllamaModel).

Step 5: Choose Output Directory

? Output directory: ./agents/weather_assistant

The default pattern is ./agents/{agent_name}. You can change this to any path.

Step 6: Review and Confirm

The wizard shows a summary:

Summary:
  Name: weather_assistant
  Description: Help users check weather forecasts...
  Model: haiku-latest
  Output: ./agents/weather_assistant

? Create this agent? Yes

Review the details and confirm to start generation.

Step 7: Wait for Generation

Creating agent 'weather_assistant'...
  Designing tools...
  Generating tool implementations...
  Creating configuration files...
  Writing files...
Done!

The creator:

  1. Analyzes your description to determine needed tools
  2. Designs tool parameters and implementations
  3. Creates a system template with appropriate instructions
  4. Generates config and task files
  5. Writes everything to the output directory

Step 8: Run Your New Agent

After generation completes, the wizard asks if you want to run the agent immediately:

Agent created successfully!

? Run the agent now? Yes

If you select Yes, the wizard will prompt you for any required task parameters and then start the agent. No need to copy-paste commands.

To run the agent later:

hugin run --task main --task-path ./agents/weather_assistant

Understanding the Generated Files

The creator produces a standard agent directory:

weather_assistant/
├── configs/
│   └── weather_assistant.yaml    # Agent configuration
├── tasks/
│   └── main.yaml                 # Default task
├── templates/
│   └── weather_assistant.yaml    # System prompt
└── tools/
    ├── get_weather.py            # Tool implementation
    ├── get_weather.yaml          # Tool definition
    ├── get_forecast.py
    └── get_forecast.yaml

Customizing the Generated Agent

The generated files are a starting point. Common customizations:

Change the System Prompt

Edit templates/{name}.yaml to adjust personality or add instructions:

template: |
  You are a friendly weather assistant.

  Always include temperature in both Celsius and Fahrenheit.
  Be encouraging about outdoor activities when weather permits.
  ...

Modify Tool Behavior

Edit the Python files in tools/ to change how tools work:

def get_weather(stack, city: str) -> str:
    # Replace simulated data with real API calls
    response = requests.get(f"https://api.weather.com/{city}")
    return response.json()

Add More Tools

Create new .py and .yaml pairs in tools/, then add them to the config:

tools:
  - builtins.finish:finish
  - get_weather:get_weather
  - get_forecast:get_forecast
  - new_tool:new_tool  # Add your new tool

Change the Default Task

Edit tasks/main.yaml to change the initial prompt or add parameters:

name: main
description: Check weather for a location
parameters:
  location:
    type: string
    description: City to check weather for
    required: true
prompt: |
  Check the weather for  and give me a summary.

Next Steps