blowsh

Use the blowsh_search_web tool, which queries DuckDuckGo, Bing, Brave, and Mojeek with cross-engine consensus. It supports intent modes like code, paper, news, and entity, plus up to two parallel query variants merged into one ranked result list.

Call blowsh_fetch_web with the URL and a type of markdown, text, HTML, or PDF. The server renders the page in a browser tier when plain HTTP fails, and options like focus, toc, and selector cut token usage by 50 to 80 percent.

Yes, blowsh_fetch_web_batch reads up to 10 URLs at once and reuses the render cache. Each item returns its own ok or error status, but batch mode drops single-fetch options like focus, toc, archive, and stitch.

Yes, blowsh_crawl_web keeps respect_robots and same_host enabled by default. You control scope with max_pages up to 200, max_depth, character budgets, include and exclude path globs, and a deadline in seconds.

On deadline expiry the tools return a deadline.hit marker or partial results rather than empty output. For 404s or timeouts, the archive option can rescue the page from the Wayback Machine when a snapshot exists.

Avoid batch mode when you need focus filtering, section targeting, archive rescue, or stitching, since fetch_web_batch silently drops those options. Use serial fetch_web calls for pages requiring those features.