Lingtai-AI/lingtai

What it solves

LingTai addresses the limitation of "one-shot" AI agents that forget context once a session ends or a terminal is closed. It provides a "Digital Scientist"—a long-lived, lifelong agent that persists within a project, accumulating knowledge, skills, and a specific operating style over weeks or months to handle complex, long-running research or engineering tasks.

How it works

LingTai is filesystem-native, storing all agent state (memory, knowledge, skills, and logs) in a local .lingtai/ directory. This makes the agent's growth inspectable and recoverable. It employs a process called "molting" to manage context windows: when the window fills, the agent crystallizes essential information into durable state and resets. This state is categorized into:

• Knowledge: A private library of research and findings.
• Skills: Reusable procedures for future tasks.
• Character: The agent's evolving expertise and goals.
• Avatars: Persistent specialist agents spawned to solve specific sub-problems.

Who it’s for

It is designed for researchers and engineers who need an AI partner capable of holding a complex question or codebase over a long period, conducting evidence-based experiments, and maintaining a durable record of findings.

Highlights

• Lifelong Memory: Durable state survives restarts and session closures.
• Filesystem-Native: State is stored in local files, allowing users to audit and edit the agent's growth.
• Multi-Interface Access: Interact via a native macOS app, a Terminal User Interface (TUI), or external channels like Telegram, WhatsApp, and email.
• Agent Orchestration: Can spawn specialist avatars or use coding agents (like Claude Code) as "hands" for implementation.
• Evidence-First Approach: Prioritizes verified findings and tool use over guesswork.