aini 0.2.0

Make class instantiation easy with auto-imports

aini

Make AI class initialization easy with auto-imports.

Installation

pip install aini

Why aini?

• Simplified Initialization: Configure complex AI components with clean YAML files
• Variable Substitution: Use environment variables and defaults for sensitive values
• Auto-Imports: No need for multiple import statements
• Debugging Tools: Inspect objects with aview for better debugging
• Reusable Configs: Share configurations across projects

Core Features

Main Components

• aini(): Loads and instantiates objects from configuration files
• aview(): Visualizes complex nested objects for debugging
• afunc(): Lists available methods on an object

Usage

Autogen

Use DeepSeek as the model for the assistant agent.

from aini import aini, aview

# Load assistant agent with DeepSeek as its model - requires DEEPSEEK_API_KEY
client = aini('autogen/client', model=aini('autogen/llm', 'ds'))
agent = aini('autogen/assistant', name='deepseek', model_client=client)

# Run the agent
ans = await agent.run(task='What is your name')

# Display result structure
aview(ans)
[Output]
<autogen_agentchat.base._task.TaskResult>
{
  'messages': [
    {'source': 'user', 'content': 'What is your name', 'type': 'TextMessage'},
    {
      'source': 'deepseek',
      'models_usage <autogen_core.models._types.RequestUsage>': {
        'prompt_tokens': 32,
        'completion_tokens': 17
      },
      'content': 'My name is DeepSeek Chat! 😊 How can I assist you today?',
      'type': 'TextMessage'
    }
  ]
}

# Display agent structure with private keys included
aview(agent._model_context, inc_=True, max_depth=5)
[Output]
<autogen_core.model_context._unbounded_chat_completion_context.UnboundedChatCompletionContext>
{
  '_messages': [
    {'content': 'What is your name', 'source': 'user', 'type': 'UserMessage'},
    {
      'content': 'My name is DeepSeek Chat! 😊 How can I assist you today?',
      'source': 'deepseek',
      'type': 'AssistantMessage'
    }
  ]
}

Agno

# Load an agent with tools from configuration files
agent = aini('agno/agent', tools=[aini('agno/tools', 'google')])

# Run the agent
ans = agent.run('Compare MCP and A2A')

# Display component structure with filtering
aview(ans, exclude_keys=['metrics'])
[Output]
<agno.run.response.RunResponse>
{
  'content': "Here's a comparison between **MCP** and **A2A**: ...",
  'content_type': 'str',
  'event': 'RunResponse',
  'messages': [
    {
      'role': 'user',
      'content': 'Compare MCP and A2A',
      'add_to_agent_memory': True,
      'created_at': 1746758165
    },
    {
      'role': 'assistant',
      'tool_calls': [
        {
          'id': 'call_0_21871e19-3de7-4a8a-9275-9b4128fb743c',
          'function': {
            'arguments': '{"query":"MCP vs A2A comparison","max_results":5}',
            'name': 'google_search'
          },
          'type': 'function'
        }
      ]
    }
  ]
  ...
}

# Export to YAML for debugging
aview(ans, to_file='debug/output.yaml')

Mem0

memory = aini('mem0/mem0', 'mem0')

Configuration File Format

aini uses YAML or JSON configuration files to define class instantiation. Here's how they work:

Basic Structure

# Optional defaults section for fallback values
defaults:
  api_key: "default-key-value"
  temperature: 0.7

# Component definition
assistant:
  class: autogen_agentchat.agents.AssistantAgent
  params:
    name: ${name}
    model_client: ${model_client|client}
    tools: ${tools}

# Nested components
mem0:
  class: mem0.Memory
  init: from_config
  params:
    config_dict:
      history_db_path: ${history_db_path}
      graph_store:
        provider: neo4j
        config:
          url: bolt://localhost:7687
          username: ${NEO4J_USERNAME}
          password: ${NEO4J_PASSWORD}

Variable Substitution

aini supports variable substitution with the ${var} syntax:

model_config:
  class: "openai.OpenAI"
  params:
    api_key: ${OPENAI_API_KEY}  # Uses environment variable
    model: ${model|'gpt-4'}     # Uses input parameter or default 'gpt-4'
    temperature: ${temp|0.7}    # Uses input parameter or default 0.7

Variable resolution priority:

1. Input variables (passed as kwargs to aini())
2. Environment variables
3. Default variables from the defaults section
4. Fallback values after the pipe | character

Custom Initialization Methods

By default, aini uses the class constructor (__init__), but you can specify custom initialization methods:

nested_example:
model_client:
  class: autogen_core.models.ChatCompletionClient
  init: load_component
  params:
    model: ${model}
    expected: ${expected}

Advanced Features

Raw Configuration Access

Use the araw parameter to get the resolved configuration without building objects:

# Get raw configuration with variables resolved
config = aini('openai/model_config', araw=True)
print(config)

# Get specific component configuration
model_config = aini('openai/model_config', akey='gpt4', araw=True)
print(model_config)