A Rust-native challenger to the $65M document parser

What it does

Extractous pulls text and metadata out of PDFs, Word files, spreadsheets, images, and email archives without phoning home to an API. The core is Rust, and it currently ships a Python wrapper that delegates to native code, with JavaScript and TypeScript bindings listed as coming soon. It targets data pipelines, RAG stacks, and LLM preprocessing where keeping everything local and in-process is the point.

The interesting bit

Rather than running a JVM or a sidecar service, the project compiles Apache Tika into native shared libraries using GraalVM ahead-of-time compilation and links them straight into the Rust core. That gives it Tika’s encyclopedic format support without the Java runtime tax—no GC, no VM, just native calls. It also leans on Tesseract when it needs to read scanned pages.

Key highlights

• The project’s own benchmarks claim roughly 18× faster throughput and 11× lower memory allocation than unstructured-io on SEC10 filing PDFs.
• Runs fully offline; no external services, no server processes, no cloud bill.
• Covers dozens of formats—including legacy Office, OpenDocument, EPUB, and image types—through either native Rust parsers or the GraalVM-compiled Tika backend.
• The Python binding is structured to dodge the GIL by doing the heavy lifting in Rust, though Python is the only binding available today.
• Licensed under Apache 2.0, including commercial use.

Caveats

• Only Python bindings exist right now; the promised multi-language support is still on the roadmap.
• OCR requires a separate Tesseract installation with language packs; it is not batteries-included.
• The performance and quality comparisons are published by the project’s own benchmarking repository, so treat the head-to-head claims accordingly.

Verdict

Give it a spin if you are building local document pipelines and want Rust’s threading without Python’s GIL getting in the way. If you need battle-tested, polyglot SDKs today, the incumbent ecosystem is still the safer bet.