Personalizing Your OpenClaw Assistant

Making OpenClaw Yours

OpenClaw is designed to adapt to you. Through configuration, memory, and skills, you can shape how your assistant communicates and behaves.

System Prompt Customization

The system prompt is the core instruction that defines your assistant's behavior. Customize it in openclaw.json:

{
  "system_prompt": "You are a helpful, concise assistant. You prefer bullet points and short answers. You have a dry sense of humor."
}

Personality Settings

Tone

• Professional: "Respond formally with complete sentences"
• Casual: "Be friendly and conversational, use contractions"
• Technical: "Assume expertise, use jargon freely, skip basic explanations"

Verbosity

• Concise: "Keep responses under 3 sentences unless asked for more"
• Balanced: "Provide thorough but focused responses"
• Detailed: "Always explain your reasoning and provide examples"

Style

• Direct: "Get straight to the point, no preamble"
• Socratic: "Ask clarifying questions before answering"
• Structured: "Use headings, bullet points, and numbered lists"

Language Preferences

{
  "language": "en",
  "response_language": "match_input"
}

Options:

• Fixed language: Always respond in a specific language
• Match input: Respond in whatever language the user writes in
• Multilingual: Mix languages as appropriate

Per-Channel Personality

Different contexts may call for different styles:

{
  "channels": {
    "slack": {
      "system_prompt_extra": "You are in a work Slack channel. Be professional and focused."
    },
    "whatsapp": {
      "system_prompt_extra": "This is a personal chat. Be casual and friendly."
    }
  }
}

USER.md Personalization

Fill out USER.md to help OpenClaw tailor responses:

• Your expertise level in different areas
• Communication preferences
• Relevant background information
• Goals and current projects

Dynamic Personalization

OpenClaw learns over time through:

• Conversational cues you provide
• Explicit "remember this" commands
• Patterns in how you interact
• Feedback on responses ("that was too long", "perfect")