The Apache Tika project is pleased to announce the release of Apache Tika 4.2.0. The release contents have been pushed out to the main Apache release site and to the Maven Central sync.
Apache Tika is a toolkit for detecting and extracting metadata and structured text content from various documents using existing parser libraries. Apache Tika 4.2.0 includes improvements for security hardening and operational safety. Configuration sent to tika-server with a request is now untrusted unless the operator marks it otherwise, and per-request limits and timeouts may only tighten the operator's. tika-server gains transfer limits (408/429 on slow or oversized requests) and a reworked TLS configuration. XML parsing moves to Apache Commons Secure XML. This version also includes bug fixes and other improvements, including docx equations and bookmarks, PE file icons, BigTIFF and RAW camera previews, and faster parser loading and fork startup. Unpack output: as announced with 4.1.0, unpacking embedded files (tika-server /unpack, tika-app -z/-Z, pipes UNPACK) now produces a Frictionless Data Package by default: a datapackage.json manifest describing each extracted file, an rmeta.json with the metadata for the container and every embedded document, and the files themselves under unpacked/. A document without attachments is still a package, so /unpack returns 200 rather than 204. To keep the previous flat layout, set unpack-config.outputFormat=REGULAR (--unpack-format=REGULAR in tika-app); unpack-config.emptyPackage=SKIP restores the 204. See: https://tika.apache.org/docs/4.2.x/pipes/unpack-config.html OCR and embeddings: the configuration changes deprecated in 4.1.0 are now enforced. OCR engines (Tesseract, Tess4J) listed under "parsers" fail at config load; configure them under "engines" and name them in "text-recognizers". A VLM parser (Claude, Gemini, OpenAI-compatible) may still sit under "parsers" to process whole documents. The openai-embedding-filter, jina-embedding-filter and openai-image-embedding-parser are removed; use an engine with an inference binding instead. New image checks keep blank and tiny images away from text recognizers, VLM output now records warnings, skip reasons and image sizes, and VLM parsers retry dropped connections. Batched recognition is planned for 4.3 as an opt-in addition; the inference Java APIs remain experimental. See: https://tika.apache.org/docs/4.2.x/configuration/inference.html https://tika.apache.org/docs/4.2.x/configuration/inference-recipes.html Please read "Breaking Changes and Deprecations" on the notable changes page before upgrading: https://tika.apache.org/4.2.0/index.html Details can be found in the changes file: https://www.apache.org/dist/tika/4.2.0/CHANGES-4.2.0.txt and in our 4.x docs site: https://tika.apache.org/docs/4.2.x/ Apache Tika is available on the download page: https://tika.apache.org/download.html Apache Tika will be available shortly in binary form or for use with Maven from the Central Repository: https://repo1.maven.org/maven2/org/apache/tika/ When downloading, please remember to verify the downloads using signatures found: https://www.apache.org/dist/tika/KEYS For more information on Apache Tika, visit the project home page: https://tika.apache.org/ Many, many thanks to fellow devs and our larger community! -- Tim Allison, on behalf of the Apache Tika community
