# Linghun

A local-first, evidence-first AI coding terminal that constrains a language model to read real code, verify its own changes, and disclose uncertainty instead of asserting that work is complete without proof.

Linghun connects a large language model to a real project's files, tools, permissions and verification commands, treating engineering conclusions as untrustworthy unless backed by evidence. It distinguishes between local verification, mocked verification, real smoke testing and unverified claims, and enforces a final-answer gate before the model can call something done.

It supports reading project structure to locate issues, editing files and generating patches, running builds, tests and scripts, checking git status and creating rollback points, indexing code to avoid repeated file reads, splitting long tasks into observable steps or background work, switching models per role such as planning, execution or review, and connecting external capabilities through MCP, skills, plugins and hooks. It runs on Windows, including PowerShell and paths with non-English characters or spaces, and installs as an npm package. Its pre-flight engine currently covers TypeScript/TSX, Python, Rust, Go and Java.

Website: https://github.com/linghungegeg/Linghun