wtf is webcmd?

Webcmd is infrastructure that helps AI agents use websites without repeating the same slow, expensive browsing every time. Normally, an AI agent that needs to check a site has to click around and explore it fresh on every run, which burns a lot of tokens and time. Webcmd tries to fix that by remembering what it learns about a site and turning that knowledge into reusable, deterministic commands. It works in four stages. First, for a site the agent has never seen, Webcmd gives it live browser control to inspect, click, type, and extract information like a person would. Second, once a site is familiar, Webcmd builds a kind of map of that site's pages, actions, and known issues, so future runs do not have to rediscover them. Third, once the useful actions are known, someone can write a proper reusable command, called an adapter, for that specific site so agents skip navigation entirely. Fourth, that adapter can be extended further so the workflow runs instantly with minimal token use. The tool is installed globally through npm and requires Node.js 20 or newer. After installing, you add its skills to your agent tool, such as Codex or Claude, and then simply describe in plain language what you want done, like researching a topic across Hacker News and Reddit or checking prices on Grainger. Webcmd already has example workflows for research sites like Hacker News, Reddit, and PubMed, social platforms like X, LinkedIn, and TikTok, AI tools like ChatGPT, Claude, and Gemini, and shopping or booking sites like Amazon and Zepto. It can also work with logged in browser sessions for tasks that need an account, such as checking LinkedIn messages or X bookmarks. There is also a hosted option called Webcmd Cloud for running these commands and browser sessions remotely, though the README notes it is still in active development and not yet stable. The project accepts community made plugins and is released under the Apache 2.0 license.