# vetd

> vetd is an open-source TypeScript checker for coding agents. The agent sends a diff. vetd checks it in memory with TypeScript 7 (tsgo), runs Next.js and React rules, and lists affected Vitest tests. It only rechecks the files the edit affects.

Requirements: Node.js 22.18+, `@typescript/native-preview` (TypeScript 7), macOS or Linux. Not on npm yet: install from a local checkout. License: Apache-2.0. Keep a full typecheck and build in CI.

Every page has a Markdown version: add `.md` to the URL (`/docs` -> `/docs.md`, `/` -> `/index.md`), or request the normal URL with `Accept: text/markdown`. All pages in one file: [llms-full.txt](http://localhost:4173/llms-full.txt).

## Docs

- [Overview](http://localhost:4173/index.md): what vetd is, how it decides what to recheck, headline benchmarks
- [Docs](http://localhost:4173/docs.md): requirements, setup, agent loop, MCP tools, check result fields, CLI, Node SDK, configuration, limits
- [MCP setup](http://localhost:4173/mcp.md): Claude Code, Cursor, Codex, VS Code, Windsurf, Gemini CLI, and the text `vetd init` adds to AGENTS.md
- [Fast type checking](http://localhost:4173/fast-typecheck.md): why running `tsc` after every agent edit is slow, and vetd compared with `tsc`

## Rules

- [Rules](http://localhost:4173/rules.md): every Next.js and React rule vetd checks, with why and how to fix
- One page per rule at `/rules/<rule id>.md`, for example [/rules/next/server-hooks.md](http://localhost:4173/rules/next/server-hooks.md)
- [Credits](http://localhost:4173/credits.md): the people and teams whose work the rules are based on

## Benchmarks

- [Benchmarks](http://localhost:4173/benchmarks.md): replayed commits on public repositories, full tsgo times, independent checks and failures
- [Raw benchmark data](http://localhost:4173/benchmarks.json)

## Optional

- [Lab](http://localhost:4173/lab): about 470 live bad-versus-good UI demos, grouped by author or topic (HTML only)
- [Sitemap](http://localhost:4173/sitemap.xml)