Intuition-Lab/personal-model

What it solves

Personal Model provides a local-first, long-term memory layer for AI coding agents. It solves the problem of agents lacking a consistent, personalized context of a user's goals, priorities, and working patterns across different tools, while ensuring that this sensitive personal data remains private and under the user's control on their own machine.

How it works

It runs as a local Runtime on macOS, capturing focused activity from apps after the user grants permissions. It organizes this information into a hierarchical structure of evidence-linked context:

• Points: Sourced observations or events.
• Lines: Relationships or changes over time.
• Faces: Patterns supported by evidence.
• Volumes: Higher-order structures across projects.
• Root: The final integrated model of the user.

This data is made available to AI clients (like Claude Code, Cursor Agent, and Codex) via the Model Context Protocol (MCP), allowing agents to search for unfinished work, rank steps against priorities, and make grounded decisions based on "receipts" rather than hidden summaries.

Who it’s for

Developers and power users on macOS who use AI coding agents and want a unified, private, and evidence-based memory that persists across different AI tools.

Highlights

• Local-first and Private: Data is stored and processed on the Mac, avoiding hosted memory services.
• MCP Integration: Works with a wide range of MCP-compatible clients including Claude Code, Cursor, and Codex.
• ** uma Unified Memory**: One "Root" model that serves as a single source of truth for multiple different AI agents.
• Evidence-Linked: Maintains provenance for claims, allowing users and agents to inspect and verify the source of a memory.