Changelog
What changed.
Newest first. We list measured results and what failed alongside what shipped.
Updates.
2026-11-06
Hosted endpoints for documents, OCR, PII, corpus, localisation and computers
The Webfold API gains document conversion, PDF OCR, PII redaction and scan, corpus redact and build, translation checks and string extraction, and sandboxed computers, with matching TypeScript, Python and Rust SDK namespaces. The API playground runs them.
2026-11-05
Documentation and API reference published
Sixteen guides and reference pages for the engine and seven for the hosted API, covering scrape, crawl, search, documents, WARC, CDP, MCP, the CLI, security and architecture.
2026-11-03
Verification run
Full engine test suite: 2,374 passed and 1 failed after fixes. Library tests 1,464 passed. A 1,928-attempt protocol suite passed in full, three runs in a row.
2026-10-15
Trained main-content selection
A small trained model now picks the blocks that are the main content. Held-out test F1 on the WCXB benchmark rose from 0.815 to 0.861.
2026-10-14
First public-benchmark result
Webfold scored on WCXB, 2,008 human-reviewed pages, with the benchmark's own scoring tool.
2026-10-11
Replay and signed requests
A replay command joins archived pages with their extracted text into one audit file. A bot-identity mode signs requests so sites can verify who is fetching.