Repository navigation
OpenSERP v0.8.20 - November 17, 2026
PDF text extraction, better Google redirect recovery, DuckDuckGo tracker blocking, and config files in binary releases.
Highlights
- PDF extraction: PDF URLs now return readable text in
fastandautoextraction modes. Markdown output joins wrapped prose. PDFs without a text layer return an error; no OCR is performed (#90). - Google: resolves
/gotolinks that return HTTP 200 HTML with a meta refresh before redirecting to the destination. Chained plaintext/urllinks are decoded locally. Destination pages are never fetched, and ad links are skipped. Improves recovery for #86; some links still remain wrapped. - DuckDuckGo:
block_trackers: truenow blocksimproving.duckduckgo.comand its subdomains without blocking search hosts (#83). - Binary releases: all eight Linux, macOS and Windows archives now include the current
config.yamlbeside the executable (#81).
Upgrade notes
- PDF extraction requires
mutoolfrom MuPDF onPATH(apt install mupdf-toolson Debian/Ubuntu). The Docker image includes it. PDFs exceedingextract.max_bytesfail explicitly instead of returning partial text. - The default
extract.max_bytesincreases from2000000to3145728bytes (3 MiB). Explicit values in existing configs still apply. - Google resolution uses at most three Google requests per wrapped link within the existing timeout budget.
resolve_urls=falseand top-N limits keep working. HTTP 400 responses and challenges can still leave links unresolved; #86 remains open. - Back up a custom
config.yamlbefore unpacking a binary release into an existing installation; archives now contain the sample configuration.
Thanks to @ramgalin for #90 and @krustfurt for #81, #83 and #86.
Full Changelog: v0.8.19...v0.8.20