knockoutez/wigolo

wigolo gives an AI coding assistant a single set of tools for everything web related: searching, fetching pages, crawling sites, extracting structured data, caching results, finding similar content, and running research tasks on its own. It is designed to run locally on your own machine, next to tools like Claude Code, Cursor, Codex, Gemini CLI, VS Code, Windsurf, Zed, and others, or inside your own self hosted agent setup. The core tools work with no API keys and no cloud service, and nothing you search or fetch leaves your own computer's storage folder. Getting started is a single command, npx wigolo init, which downloads the local browser engine and on device models, checks that everything works, and reports the status of each piece. You can also tell it which coding agents to wire up automatically in that same command, so a tool you already use daily is ready right away. It needs Node version 20 or newer and about 1.5 gigabytes of free disk space, and works on macOS, Linux, and Windows. Search, fetch, crawl, extract, cache, and find similar all work without any paid key. Two more advanced tools, research and agent, use a language model to write a finished, cited answer instead of just returning raw search results, so those work better with a free API key from a provider like Gemini, though other providers or a fully local model work too. Every search result comes back with an exact excerpt from the source page, a citation the agent can quote, and a score showing how trustworthy that result is. Weak or unreliable results get flagged automatically, and any cached or stale data is labeled as such, so the agent always knows what it is working with. wigolo is offered under the AGPL 3.0 license and is currently in public beta.