The Rust crate is pdf-inspector — the library's native form. Every other binding (Node/Python/WASM) compiles from it, so in Rust you get ceiling performance. The default build is pure Rust with no ML models; lopdf is the only PDF parsing dependency.

Install

cargo add pdf-inspector

For unreleased changes:

[dependencies]
pdf-inspector = { git = "https://github.com/firecrawl/pdf-inspector" }

Core APIs

use pdf_inspector::{process_pdf, detect_pdf, PdfType};

fn main() -> Result<(), Box<dyn std::error::Error>> {
    // 1. All-in-one: detect + extract + Markdown
    let result = process_pdf("document.pdf")?;
    println!("Type: {:?}", result.pdf_type);       // TextBased / Scanned / ImageBased / Mixed
    println!("Confidence: {:.0}%", result.confidence * 100.0);
    if let Some(markdown) = &result.markdown {
        println!("{markdown}");
    }

    // 2. Classification only (no extraction)
    let info = detect_pdf("document.pdf")?;
    match info.pdf_type {
        PdfType::TextBased => { /* extract locally */ }
        _ => { /* info.pages_needing_ocr tells you which pages */ }
    }
    Ok(())
}

Byte input: process_pdf_mem(&bytes) — no filesystem needed.

Fine-grained control with PdfOptions

use pdf_inspector::{process_pdf_with_options, PdfOptions, ProcessMode, DetectionConfig, ScanStrategy};

// Analyze layout without generating markdown
let result = process_pdf_with_options(
    "document.pdf",
    PdfOptions::new().mode(ProcessMode::Analyze),
)?;

// Custom detection strategy (sample 5 pages)
let result = process_pdf_with_options(
    "large.pdf",
    PdfOptions::new().detection(DetectionConfig {
        strategy: ScanStrategy::Sample(5),
        ..Default::default()
    }),
)?;

// Selected pages only
let result = process_pdf_with_options(
    "document.pdf",
    PdfOptions::new().pages([1, 3, 5]),
)?;

ProcessMode has three levels: DetectOnly, Analyze, and Full.

CLI tools

Two binaries ship with the crate — cargo install pdf-inspector gets you both:

# pdf2md: PDF → Markdown
pdf2md document.pdf                 # stdout
pdf2md document.pdf --json          # structured JSON
pdf2md document.pdf --pages 1-3     # selected pages
pdf2md document.pdf --compact       # token-saving output

# detect-pdf: classification
detect-pdf document.pdf --analyze --json

Optional: OCR capability

The native build can opt into selective OCR via feature flags without pulling an inference runtime:

[dependencies]
pdf-inspector = { version = "1", features = ["vision", "model-cache"] }

• The vision feature provides the PageRenderer / OcrEngine traits and OcrOptions (Off / Auto / Force routing)
• OCR defaults to off — clean processing requests never touch models or the network
• model-cache + model-download handle pinned PP-OCRv6 Small artifacts with checksum verification and atomic installation; offline consumers set an explicit model directory with ModelDownloadPolicy::Offline

Full contracts in the official docs/rust-api.md and the OCR runtime guide.

Tips

• Don't call process_pdf when you only need the verdict — detect_pdf is an order of magnitude faster
• Lower the sampling count for large files (ScanStrategy::Sample(n)) to speed up classification noticeably
• The official docs/rust-api.md is the source of truth