Give any local model live web access — search the web and read pages — as callable tools.
Part of lmstudio-suite.
| Tool | What it does |
|---|---|
web_search | Search the web; returns a ranked list of results (title, URL, snippet). |
fetch_url | Fetch an http/https page and return its main content as clean Markdown. |
http_request | Call a REST API (GET/POST/PUT/PATCH/DELETE) with headers + body; byte-capped. |
download_file | Download a file into the configured download directory (size-capped). |
crawl | Follow same-origin links from a start URL (bounded depth + pages) → Markdown. |
Typical flow: web_search to find sources, fetch_url to read one; http_request hits JSON APIs, crawl reads a small doc section, download_file saves an asset. Every request blocks private/loopback hosts by default (SSRF), re-validated across redirects.
Global (app-wide):
DuckDuckGo (no key, default), Tavily, Brave, or SearXNG.Per-chat:
Enable web-tools in a chat with a tool-capable model, then ask something that needs current information — the model will call web_search / fetch_url on its own.
MIT licensed.