Summary

Persome builds a "Personal Model" (HUMAN.md) on macOS by learning from user activity, enabling AI clients like Codex and Claude Code to access evidence-linked context. This open-source project transforms AI agents into personalized, proactive assistants that understand user priorities and decision-making styles, running privately and locally. It targets Mac users seeking more grounded and context-aware AI interactions.

How It Works

Persome constructs a hierarchical context model from user activity: Points -> Lines -> Faces -> Volumes -> Root. This approach provides deterministic "Runtime proof" and ensures AI inferences are supported by verifiable "receipts" or sourced evidence. The model dynamically adapts to new data, offering a novel way for AI to tailor responses to individual users. Local OCR uses PP-OCRv6 (Apple Silicon) or Apple Vision (Intel).

Quick Start & Requirements

Requires macOS 13+ and Xcode Command Line Tools. A 5-minute synthetic demo uses git clone https://github.com/Intuition-Lab/personal-model.git && cd personal-model && uv run python scripts/sample_demo.py, requiring only Git and uv. For personal data installation: uv tool install personal-model then persome onboard and persome model open --after 30, or clone and run bash install.sh. Onboarding requires Accessibility/Screen Recording permissions. LLM integration is optional (persome llm setup). MCP clients install via persome install [client-name].

Highlighted Details

• Local & Private: Operates entirely on the user's Mac, ensuring data privacy and inspectability.
• Personalized AI Context: Enables AI agents to understand individual user goals and habits for more relevant outputs.
• Evidence-Linked Reasoning: AI claims are traceable to sourced observations ("receipts"), reducing fabrication.
• MCP Client Integration: Supports seamless connection with trusted AI clients.

Maintenance & Community

Contributors include Singularity (Lead), Li_Xufeng, Siyi, Kevin, huachenjie238-oss, Jing@Meowy, and Zhiheng Chen. No specific community channels or roadmap links are provided.

Licensing & Compatibility

The specific open-source license is not stated in the README. Compatibility is limited to macOS 13+.

Limitations & Caveats

Strictly macOS only. Requires significant system permissions (Accessibility, Screen Recording). Potential early-stage development indicated by onboarding focus and non-guaranteed recall times. The absence of a stated license prevents assessment of commercial use compatibility.