Browses a compressed archive — "open / list / what is in this .zip, .tar.gz, .7z, .rar", then "extract member X from it". Not a web page fetcher — a URL to a page goes to fetch. Call archive{source:"https://…/data.zip"} for the member listing (names and sizes, nothing extracted); then archive{source:"…", member:"docs/readme.md"} converts that one member to Markdown. 0 members = an empty archive; an ERROR item = the archive could not be fetched or opened, or the member was refused (traversal, symlink, size).

Retrieves 1–50 documents as Markdown, input order kept. Call fetch{sources:["https://…","doi:10.…","harvest:…"]} — a URL/path, DOI, ISBN, PMID/PMCID, or a handle from findWorks/searchCache; a title → findWorks first. Each item returns content (may be truncated), size, cache status, and the path to the COMPLETE artifact — read the path for the rest. A failing item carries its own error and the others still return; an empty extraction is reported as that item's error, never as a blank body.

Fetches images as local files for vision — "get this figure / photo / scanned page". Call fetchImage{sources:["https://…/fig1.png"]}, 1–50 URLs, local paths, or harvest handles, results in input order. Returns each item's path and bytes — open the path with vision; images are never OCR'd or converted. A failing item carries its own error and the rest still return. For a document, PDF, or web page use fetch.

Finds scholarly papers and books by TITLE or bibliographic query — "find the paper about X", "is there a PDF of ". No download. Call findWorks{query:"Attention Is All You Need"}. Returns ranked candidates (title, authors, year, kind, access) each with a fetch handle — pass that value unchanged to fetch. Empty candidates = nothing matched (give the exact title); a tool error = discovery itself failed, retry or fetch an exact identifier. For a general web query use search.

Greps the local cache of already-fetched documents — "did we already fetch X", "which cached pages mention Y". Call searchCache{pattern:"transformer attention"}. Returns WHICH cached pages match (url, md_path, match count, sample line), not their text — read md_path or fetch the url for content. Empty matches = nothing held matches (the web is not searched; search or fetch first); a tool error = the pattern was invalid or the cache could not be read.