An agent that models your identity before your inbox

What it does

Elephant Agent maintains a correctable Personal Model of the user—identity, surroundings, current priorities, and accumulated life lessons—then uses that understanding to design long-running Paths for work, health, learning, or relationships. It exposes all of this through a native macOS app or a CLI-plus-web-dashboard setup for Linux and cloud machines, keeping the model inspectable instead of buried in chat history.

The interesting bit

The project explicitly ranks itself as an L4 agent that grows with the person, above agents that merely execute tasks or carry context. It organizes its user model into four lenses—Identity, World, Pulse, and Journey—and delegates steps to “baby elephants” while a “Mother Elephant” maintains the overarching understanding.

Key highlights

• Persistent Personal Model split across Identity, World, Pulse, and Journey
• Long-running Paths broken into steps with checkpoints and sub-agent delegation
• Native macOS app alongside a CLI and web dashboard for Linux/cloud use
• Inspectable token usage, runtime state, and model provider configuration
• Messaging integrations for WeChat, Discord, Feishu, DingDing, and WeCom

Caveats

• The README stays product-first and avoids implementation details; how the Personal Model is technically stored or queried is unclear
• “Self-evolving” is promised but the mechanics of evolution aren’t explained in the source
• macOS is the recommended surface, suggesting the Linux/CLI experience may be less polished

Verdict

Give it a spin if you want an AI that maintains a persistent, inspectable model of your preferences and long-term goals. Look elsewhere if you need a battle-tested framework with deep technical documentation, or if anthropomorphic agent metaphors aren’t your thing.