PageSift Changelog
Release history
One highlight block per release, from the repository's changelog and release notes. The repository also keeps the full changelog back to 0.01.
1.2.2
Nov 8, 2026Latest- Owner-scoped jobs — crawl, batch and extract jobs belong to their owner; only the owner can read, stream, list or cancel them
- Page quotas — page fetches for search and extract count against the daily page quota
- Cache respects robots.txt — cached results are re-checked against robots.txt before they are served
- Corrections — unsupported snapshot and base64-image options return an error instead of being ignored; MCP placeholder outputs match their schemas
1.2.1
Nov 8, 2026- Cache isolation — requests carrying an identity session never use the shared result cache, so one tenant cannot receive another tenant's session-scoped response
- Throttling classified — 503 with Retry-After is recorded as throttled in the itna attempt trace
- MCP schemas — every tool gains an output schema and complete parameter descriptions
- Tool changes —
pagesift_crawl_statusbecomespagesift_poll_job; newpagesift_cancel_jobstops running jobs
1.2.0
Nov 7, 2026- itna — access-barrier recovery, opt-in and off by default: classify the barrier first, then spend at most one browser fallback on it
- Owner- and origin-scoped sessions — identity leases by owner, origin, egress and fingerprint; requests with credentials or custom headers neither share a session nor escalate
- Validated proxy tunnels for the TLS fingerprint, with a restricted HTTP-to-browser cookie handoff (exact origin, SameSite=Strict)
- Bounded waits for non-interactive challenges; the automation marker is removed, no hardware identity is fabricated, and interactive CAPTCHAs return
needsHuman— no CAPTCHA solver is included
1.1.3
Nov 6, 2026- Pure-Rust zstd compression and decompression updates
- Cache-aware matching and Huffman tables
- Corrected repeated offsets when falling back to raw blocks
1.1.2
Nov 4, 2026- The MCP server implements
resources/listandresources/templates/list - Verilog source files are split by module
- Per-release PGO profile data ships with the release
1.1.1
Nov 4, 2026- Source-code sections for 30 languages
- .doc gains list numbers, headers, footers and comments
- dBASE memo files inside archives
pagesift-compressreplaces the C zstd library — compression is pure Rust
1.1.0
Nov 4, 2026- New input formats — .doc, .xls, .ppt, .xlsb, OpenDocument including flat XML, RTF, WordPerfect, HWP and HWPX, MOBI, FB2, email (.eml, .mbox, .msg), recursively read archives, image files through OCR, SQLite, dBASE, DocBook, JATS, SVG, OPML, subtitles and notebooks
- 373 source-code and config languages as fenced code blocks; C, C++, C#, Go, Java, JavaScript, TypeScript, Lua, PHP, Python and Rust split by function and class
- LangChain and LlamaIndex integrations
- C ABI with language bindings — Go, Java, Kotlin, C#, Ruby, PHP, Lua, Perl, Dart, Swift and Zig
1.0.0
Nov 4, 2026- OCR for scanned PDFs on CPU or NVIDIA GPU — own inference runtime and CUDA layer
- Structured extraction with selector memory and relocation
- Browser actions and sessions;
/v2/agent;/v2/monitor - Full Jina Reader API —
/r/and/s/; WARC input; hooks and[[rules]]; optional browser TLS fingerprint