Linghun provides an AGI-oriented, hallucination-resistant AI coding runtime designed to ground large language models in real projects, tools, permissions, and verification. It targets individual developers, novices, and enterprise teams seeking to move AI coding beyond simple Q&A to robust, controllable, and deliverable engineering tasks. Linghun offers a significant benefit by reducing AI-driven coding errors, rework, and context costs through its evidence-first, system-level approach.

How It Works

Linghun acts as an "engineering exoskeleton" for LLMs, managing their interaction with the real world. Its core approach is an "evidence-first engineering loop" that enforces constraints on reading facts, verifying actions, distinguishing evidence scope, and stating uncertainties. This is achieved through integrated capabilities like tool execution, permission management, evidence logging, Git integration, code indexing, and a centralized scheduling system. This "base layer interlocking capability" ensures that AI-driven conclusions are grounded in verifiable actions rather than solely model confidence.

Quick Start & Requirements

• Installation: npm install -g @linghun/cli
• Prerequisites: Node.js 22+ and a Node package manager (npm, pnpm).
• Usage: Run linghun within a project directory. Model configuration is handled via /model setup.
• Documentation: Links to the English README, Chinese and English Whitepapers, Update Logs, and App Bridge documentation are provided within the project's README.

Highlighted Details

• Achieved high rankings in the Novita x Harbor Agent Benchmark (e.g., 1st in Systems & Security, Data & Science).
• Employs an "evidence-first" methodology, grounding AI outputs in read files, tool results, Git status, and verification records.
• Features robust support for Windows environments and Chinese language interaction as first-class citizens.
• Integrates a comprehensive "architecture system" and "AntiCodeBlob" to promote code health and prevent architectural decay.
• Offers "verification-aware delivery," distinguishing between various completion states (PASS, PARTIAL, FAIL) and verification types.
• Supports extensibility through MCP, Skills, Plugins, Hooks, and the App Bridge for integrating external capabilities.

Maintenance & Community

The project is under active development with frequent updates. Community interaction is facilitated via a QQ group (1104150634). The project acknowledges support from "geek2api 中转站".

Licensing & Compatibility

Licensed under the Apache License 2.0. This license is permissive, allowing for commercial use and integration into closed-source projects.

Limitations & Caveats

Linghun is still under active development, with some advanced features like multi-platform native runner packaging, remote channel product experience, and public documentation requiring further maturation. Users are advised to treat it as a local engineering assistant with evidence and permission boundaries, rather than a fully autonomous system.