Peripheral

关于

Peripheral constructs a detailed self-model from your conversation history with Claude and ChatGPT. It systematically parses thousands of messages to extract behavioral signals such as corrections, taste preferences, directives, emotional patterns, and identity markers. These signals are then distilled into two distinct constitutions – one executive (how you work) and one emotional (who you are). The system further refines this model by interviewing you across 10 psychological dimensions, using prediction-based coverage to ensure the AI deeply understands you by anticipating your responses. The resulting self-model is then served via MCP tools, enabling any AI agent to write, decide, and review content consistent with your unique voice and preferences.

主要功能

• 0 GitHub stars
• Parse AI conversation history from Claude Code, Claude web, and ChatGPT into structured behavioral signals.
• Distill extracted signals into two constitutions: executive (work style) and emotional (personality).
• Conduct AI-driven interviews across 10 psychological dimensions using prediction-based coverage to fill model gaps.
• Serve the self-model as MCP tools: `get_self_model`, `get_style_rules`, `write_as`, and `would_approve`.
• Support for both Anthropic (Claude) and OpenAI (ChatGPT) models for distillation, interviews, and MCP tools.

使用案例

• Enable AI agents to ghostwrite content in your specific voice and style.
• Allow AI to evaluate generated work against your personal taste, values, and preferences.
• Develop a personalized AI assistant that understands and reflects your unique personality and operational workflow.