Skip to content

System Configuration

Create system.yml in ~/.config/kagura/:

system:
  language: en
prompt:
  instructions:
    - language: en
      description: |
        Your name is Kagura.
        You will always respond in English.
    - language: ja
      description: |
        あなたの名前は、神楽（かぐら）です。
        あたなたは、必ず日本語で返答します。
memory:
  message_history:
    history_uuid: kagura_personal_chat
    window_size: 1000
    context_window: 20
    ttl_hours: 24
  backend:
    default_ttl_hours: 24
    cleanup_interval_hours: 1
llm:
  model: openai/gpt-4o-mini
  max_tokens: 4096
  retry_count: 3
backends:
 - name: redis
   host: localhost
   port: 6379
   db: 0

Field Descriptions

system:

  • language: Defines the default language for system instructions. Supported values include en (English) and ja (Japanese).

prompt:

  • instructions: A list of language-specific instructions describing how the assistant should behave.
  • language: Specifies the language of the instruction.
  • description: Detailed instructions for the assistant's behavior.

memory.message_history:

  • history_uuid: Unique identifier for the chat history.
  • window_size: Number of messages to store in history.
  • context_window: Number of messages to consider as context for responses.
  • ttl_hours: Time-to-live for the chat history in hours.

memory.backend:

  • default_ttl_hours: Default time-to-live for memory storage.
  • cleanup_interval_hours: Interval in hours for cleaning up memory storage.

llm:

  • model: The LLM model to use (e.g., openai/gpt-4o-mini).
  • max_tokens: Maximum number of tokens the model can generate per response.
  • retry_count: Number of retry attempts in case of model failure.

backends:

List of backend services for storage or processing.

  • name: Name of the backend (e.g., redis).
  • host: Host address of the backend.
  • port: Port number for the backend service.
  • db: Database index to use in the backend.

Supported LLM Providers

Kagura supports multiple providers via LiteLLM:

  • OpenAI (openai/gpt-4, openai/gpt-3.5-turbo)
  • Anthropic (anthropic/claude-3-opus, anthropic/claude-3-sonnet)
  • Ollama (ollama/llama3.2, ollama/gemma2.5)
  • Google (google/gemini-pro)

For a complete list of supported providers, see the LiteLLM Documentation.

AI Agents Configuration

For a detailed explanation of agent-related settings, refer to the AI Agents Overview:

Quick Start →