Doppel — AI Digital Twins from Chat History

Create realistic AI digital twins of real people by analyzing their WhatsApp chat exports.

⚠️ Ethics & Consent

Informed consent is required. Before creating a digital twin of any person, you must obtain their explicit, informed consent. This means:

• Inform the person that their chat messages will be used to create an AI agent that mimics their personality and communication style.
• Explain what the twin will be used for (personal use, entertainment, testing, etc.).
• Obtain explicit permission — verbal or written — before proceeding.
• Respect refusal. If the person declines, do not create the twin.
• Allow withdrawal. The person may request deletion of the twin at any time.

The agent must confirm consent before proceeding. When a user requests a twin, ask: "Do you have this person's permission to create an AI twin of them?" Do not proceed unless the user confirms.

How It Works

• User provides a WhatsApp chat export (.txt file)
• Parser script extracts and categorizes messages by sender
• LLM analyzes the parsed messages to generate personality profile files
• A new OpenClaw agent is created with the twin's identity

Quick Start

Step 1: Get the Chat Export

Tell the user to export the chat from WhatsApp:

• Open the chat → ⋮ → More → Export Chat → Without Media
• Send the .txt or .zip file

Step 2: Parse the Chat

Run the parser to extract messages:

python3 scripts/parse_chat.py <chat_export.txt> <target_name> <output_dir>

Arguments:

• chat_export.txt — Path to the WhatsApp export file
• target_name — Name of the person to clone (as it appears in the chat)
• output_dir — Directory to save parsed output

Step 3: Generate Twin Profile

Using the parsed messages, create these files in the twin's workspace:

• SOUL.md — Read references/soul-guide.md for structure
• EXAMPLES.md — Read references/examples-guide.md for structure
• ANTI-EXAMPLES.md — Read references/anti-examples-guide.md for structure
• MEMORY.md — Read references/memory-guide.md for structure

• Identifying ALL recurring phrases, gírias, and expressions
• Capturing emotional patterns and tone shifts
• Noting relationship dynamics with the chat partner
• Extracting real events, people, and shared history

Step 4: Create the Agent

• Create workspace: ~/.openclaw/workspace-/
• Create agent dir: ~/.openclaw/agents//agent/
• Place SOUL.md, EXAMPLES.md, ANTI-EXAMPLES.md, MEMORY.md in workspace
• Create AGENTS.md in agent dir — read references/agents-guide.md for template
• Register agent in OpenClaw config via gateway config.patch

Step 5: Test

Send a message to the twin and verify it responds in character. Iterate on SOUL.md if needed.

Privacy & Data Handling

• Chat exports should be processed locally and never transmitted to external services beyond the LLM API call.
• Parsed data and generated profiles are stored in the user's local OpenClaw workspace.
• Users are responsible for securing twin data and respecting the subject's privacy.
• If the subject requests deletion, remove all workspace files, agent configuration, and parsed data.

Tips for Better Twins

• More messages = better twin. 50+ messages from the target person recommended.
• Text messages matter most. Audio/image/sticker messages are lost in export.
• Recent chats are better. People's communication style evolves.
• Multiple chats help. If available, combine exports from different chats with the same person.
• Iterate. Test the twin, identify gaps, refine SOUL.md.