run-llama/liteparse

LiteParse

| | | | | | Docs

English | 简体中文

Looking for LiteParse V1? Follow this link to the old code

LiteParse is a standalone OSS PDF parsing tool focused exclusively on fast and light parsing. It provides high-quality spatial text parsing with bounding boxes, without proprietary LLM features or cloud dependencies. Everything runs locally on your machine.

Hitting the limits of local parsing? For complex documents (dense tables, multi-column layouts, charts, handwritten text, or scanned PDFs), you'll get significantly better results with LlamaParse, our cloud-based document parser built for production document pipelines. LlamaParse handles the hard stuff so your models see clean, structured data and markdown.

Sign up for LlamaParse free

Overview

• Fast Text Parsing: Spatial text parsing using PDFium, ~2-5ms per page
• Flexible OCR System:
• Built-in: Tesseract (zero setup, bundled with the library)
• HTTP Servers: Plug in any OCR server (EasyOCR, PaddleOCR, custom)
• Standard API: Simple, well-defined OCR API specification
• Complexity Detection: Cheaply check whether a document needs OCR or heavier parsing — route, reject, or estimate cost before a full parse
• Screenshot Generation: Generate high-quality page screenshots for LLM agents
• Multiple Output Formats: Markdown, JSON, and Text
• Markdown Output: Structured Markdown with headings, tables, lists, images, and links — great for feeding LLMs and RAG pipelines
• Bounding Boxes: Precise text positioning information
• Multi-language: Use from Rust, Node.js/TypeScript, Python, or the browser (WASM)
• Worker Pool Mode (Python & Node.js): Parse in persistent worker processes for true parallelism (PDFium otherwise serializes concurrent parses) and hard per-parse timeouts — rogue documents are killed, identified by name, and never stall the pipeline
• Multi-platform: Linux, macOS (Intel/ARM), Windows

flowchart LR
      subgraph Input["Input Formats"]
          direction TB
          PDF["PDF"]
          DOCX["DOCX"]
          XLSX["XLSX"]
          PPTX["PPTX"]
          IMG["Images"]
      end
subgraph Core["Rust Core"]
          direction TB
          CONV["Format Conversion\nLibreOffice / Rust image + resvg + usvg crates"]
          EXTRACT["Text Extraction\nPDFium C library"]
          OCR["Selective OCR\nTesseract / HTTP / Custom"]
          MERGE["OCR Merge\nNative text + OCR results"]
          PROJ["Grid Projection\nSpatial layout reconstruction"]
          CONV --> EXTRACT
          EXTRACT --> OCR --> MERGE --> PROJ
          EXTRACT --> MERGE
      end
subgraph Output[" Output "]
          direction TB
          JSON["Structured JSON\ntext + bounding boxes"]
          TEXT["Plain Text\nlayout-preserved"]
          SCREEN["Screenshots\nPNG rendering"]
      end
subgraph Bindings["Language Bindings"]
          direction TB
          NAPI["Node.js / TypeScript\nnapi-rs"]
          PYO3["Python\nPyO3"]
          WASM["Browser / WASM\nwasm-bindgen"]
          CLI["CLI\ncargo / npm / pip"]
          NAPI ~~~ PYO3 ~~~ WASM ~~~ CLI
      end
PDF --> EXTRACT
      DOCX & XLSX & PPTX & IMG --> CONV
      PROJ --> JSON & TEXT & SCREEN
      JSON & TEXT & SCREEN --> Bindings
style Input fill:#F5F5F5,color:#000000,stroke:#37D7FA,stroke-width:2px
      style Core fill:#F5F5F5,color:#000000,stroke:#3E18F9,stroke-width:2px
      style Output fill:#F5F5F5,color:#000000,stroke:#FF8705,stroke-width:2px
      style Bindings fill:#F5F5F5,color:#000000,stroke:#FF8DF2,stroke-width:2px
style PDF fill:#96E7F9,color:#000000,stroke:#37D7FA,stroke-width:1px
      style DOCX fill:#96E7F9,color:#000000,stroke:#37D7FA,stroke-width:1px
      style XLSX fill:#96E7F9,color:#000000,stroke:#37D7FA,stroke-width:1px
      style PPTX fill:#96E7F9,color:#000000,stroke:#37D7FA,stroke-width:1px
      style IMG fill:#96E7F9,color:#000000,stroke:#37D7FA,stroke-width:1px
style CONV fill:#92AEFF,color:#000000,stroke:#4B72FE,stroke-width:1px
      style EXTRACT fill:#92AEFF,color:#000000,stroke:#4B72FE,stroke-width:1px
      style OCR fill:#92AEFF,color:#000000,stroke:#4B72FE,stroke-width:1px
      style MERGE fill:#92AEFF,color:#000000,stroke:#4B72FE,stroke-width:1px
      style PROJ fill:#4B72FE,color:#FFFFFF,stroke:#3E18F9,stroke-width:2px
style JSON fill:#FFBD74,color:#000000,stroke:#FF8705,stroke-width:1px
      style TEXT fill:#FFBD74,color:#000000,stroke:#FF8705,stroke-width:1px
      style SCREEN fill:#FFBD74,color:#000000,stroke:#FF8705,stroke-width:1px
style NAPI fill:#FFBFF8,color:#000000,stroke:#FF8DF2,stroke-width:1px
      style PYO3 fill:#FFBFF8,color:#000000,stroke:#FF8DF2,stroke-width:1px
      style WASM fill:#FFBFF8,color:#000000,stroke:#FF8DF2,stroke-width:1px
      style CLI fill:#FFBFF8,color:#000000,stroke:#FF8DF2,stroke-width:1px

Benchmarks

LiteParse is measured on multiple public Doc→Markdown benchmarks. All numbers below were produced on this machine from one command (see Reproducing), with every tool at its latest release as of 2026-09-09. LiteParse and the "model-free" competitors use no ML model at all (no LLM, no layout model, no GPU). All tested methods have permissive licenses and can be run locally with minimal dependencies.

| Benchmark | Metric | LiteParse | + Tesseract OCR | + PaddleOCR | Best other model-free tool | |---|---|---|---:|---:|---:| | ParseBench (2,049 docs) | Overall (mean of 5 categories) | 0.364 | 0.380 | 0.389 | pdf-inspector 0.283 | | opendataloader-bench (200 docs) | Overall (NID + TEDS + MHS) | 0.886 | 0.896 | 0.901 | opendataloader 0.842 | | olmOCR-bench (1,403 pages) | % tests passed | 39.6 | 41.1 | 42.2 | pdf-inspector 33.7 |

ParseBench

Rule-based scoring, no LLM judge. Each column is ParseBench's canonical per-category metric (Tables = GriTS/TRM composite; the others are rule pass-rates). Overall is the mean of the five, as on the ParseBench leaderboard.

| Pipeline | Overall | Tables | Charts | Content Faithfulness | Semantic Formatting | Visual Grounding | |---|---:|---:|---:|---:|---:|---:| | LiteParse + PaddleOCR | 0.389 | 0.430 | 0.013 | 0.787 | 0.402 | 0.314 | | LiteParse + Tesseract | 0.380 | 0.428 | 0.012 | 0.751 | 0.399 | 0.307 | | LiteParse (no OCR) | 0.364 | 0.424 | 0.013 | 0.700 | 0.385 | 0.297 | | pdf-inspector 1.19 | 0.283 | 0.277 | 0.017 | 0.598 | 0.426 | 0.099 | | opendataloader 2.5.7 | 0.277 | 0.349 | 0.006 | 0.663 | 0.258 | 0.110 | | markitdown 0.1.7 | 0.185 | 0.158 | 0.020 | 0.652 | 0.001 | 0.110 |

Notes:

• Visual Grounding scores layout blocks with bounding boxes (lit parse --extract-blocks).
• Charts is near zero for every tool here — none reconstruct chart data.

opendataloader-bench

NID = reading-order similarity, TEDS = table structure, MHS = heading hierarchy. Overall is the harness's own mean.

| Engine | Overall | NID | TEDS | MHS | |---|---:|---:|---:|---:| | LiteParse + PaddleOCR | 0.901 | 0.932 | 0.832 | 0.840 | | LiteParse + Tesseract | 0.896 | 0.928 | 0.829 | 0.828 | | LiteParse (no OCR) | 0.886 | 0.917 | 0.818 | 0.821 | | nutrient (commercial) | 0.885 | 0.925 | 0.708 | 0.819 | | opendataloader 2.5.7 | 0.842 | 0.912 | 0.483 | 0.757 | | markitdown 0.1.7 | 0.589 | 0.844 | 0.273 | 0.000 |

Notes:

• nutrient has no runnable parser in this harness (commercial)
• The corpus is native-text PDFs, so OCR gains come from embedded figures and a handful of

scanned pages.

olmOCR-bench

Score = average of per-category pass rates. The two math categories require LaTeX output and are 0% for every tool here; they still count in the average.

| Engine | Overall | baseline | headers_footers | multi_column | table_tests | long_tiny_text | old_scans | arxiv_math | old_scans_math | |---|---:|---:|---:|---:|---:|---:|---:|---:|---:| | LiteParse + PaddleOCR | 42.2 | 99.9 | 48.7 | 69.1 | 54.0 | 46.4 | 19.4 | 0.0 | 0.0 | | LiteParse + Tesseract | 41.1 | 99.9 | 52.1 | 66.2 | 54.1 | 42.5 | 13.9 | 0.0 | 0.0 | | LiteParse (no OCR) | 39.6 | 99.9 | 55.8 | 66.3 | 52.5 | 29.2 | 13.3 | 0.0 | 0.0 | | pdf-inspector 1.19 | 33.7 | 82.9 | 62.1 | 49.7 | 43.6 | 17.6 | 13.3 | 0.0 | 0.0 | | opendataloader 2.5.7 | 32.5 | 86.9 | 36.6 | 63.7 | 24.9 | 34.8 | 13.3 | 0.0 | 0.0 | | markitdown 0.1.7 | 28.7 | 86.8 | 38.8 | 39.3 | 19.9 | 31.2 | 13.3 | 0.0 | 0.0 |

Notes:

• headers_footers expects letterhead and footer text to be absent. OCR recovers that text

from logo and address images on single-page documents, where the repeated-header filter cannot fire, so the OCR rows score lower there by design. We chose not to drop text by page position. pdf-inspector 1.19 refuses scanned and image-based pages outright (192 of the 1,403), which is why it tops this category while scoring nothing on those pages elsewhere.

• old_scans is largely cursive handwriting; Tesseract cannot read it, PaddleOCR partially can.

How the runs are configured

• Same scorer for every tool. Each benchmark's own evaluator, unmodified. The ground truth in

all three corpora is plain text, so every tool including LiteParse runs with hyperlink syntax off (--no-links); everything else is the default lit parse --format markdown.

• OCR modes. No OCR is --no-ocr. Tesseract is the built-in engine with no setup.

PaddleOCR is the PP-OCRv5 mobile detector and English recognizer served over the OCR HTTP API by ocr/rapidocr, which runs the PaddleOCR models on ONNX Runtime — the same models as ocr/paddleocr, just 20–40× faster per page on CPU-only machines.

• Competitors are the free converters we could run locally, each at its latest release on

2026-09-09: markitdown 0.1.7, opendataloader-pdf 2.5.7 (its non-hybrid mode), pdf-inspector 1.19.0. Each benchmark's own runner for the tool is used as-is. Numbers from public leaderboards or LLM-assisted modes are not mixed in.

• Machine: Apple M2 Max, 12 cores, 32 GB. LiteParse rows: v2.14.4 (2026-09-09).

Reproducing the benchmarks

./run_benchmarks.sh --liteparse-only                # all three benches, no OCR
./run_benchmarks.sh --liteparse-only --ocr=tesseract
( cd ocr/rapidocr && uv run server.py ) &           # then:
./run_benchmarks.sh --liteparse-only --ocr=paddle
./run_benchmarks.sh --competitors-only              # re-run the free competitors only
./run_benchmarks.sh                                 # everything

Results and a SUMMARY.md land in bench_results/latest/.

Installation

Install via your preferred package manager. All versions (except WASM) ship with the same lit CLI.

| Language | Install | Library Docs | |----------|---------|--------------| | Node.js / TypeScript | npm i -g @llamaindex/liteparse | Node.js README | | Python | pip install liteparse | Python README | | Rust | cargo install liteparse (CLI) / cargo add liteparse (lib) | Rust README (crates.io) | | Browser (WASM) | npm i @llamaindex/liteparse-wasm | WASM README |

Agent Skill

You can use liteparse as an agent skill, downloading it with the skills CLI tool:

npx skills add run-llama/llamaparse-agent-skills --skill liteparse

Or copy-pasting the SKILL.md file to your own skills setup.

See the Agent Skill guide for requirements and usage patterns.

CLI Usage

The CLI is the same across all installations (npm, pip, cargo install).

Parse Files

# Basic parsing
lit parse document.pdf
Parse to Markdown — headings, tables, lists, images, links
lit parse document.pdf --format markdown -o output.md
Parse with specific format
lit parse document.pdf --format json -o output.json
Parse specific pages
lit parse document.pdf --target-pages "1-5,10,15-20"
Parse without OCR
lit parse document.pdf --no-ocr
Include page-scoped vector path data in JSON
lit parse document.pdf --format json --extract-vector-graphics
Include rich per-item PDF text metadata
lit parse document.pdf --format json --extract-text-metadata
Include page annotations in structured JSON
lit parse document.pdf --format json --extract-annotations
Include AcroForm widget fields and values (repairs orphaned widgets in memory)
lit parse document.pdf --format json --extract-form-fields
Parse a remote PDF
curl -sL https://example.com/report.pdf | lit parse -

Markdown Output

LiteParse can render documents directly to Markdown. This means reconstructing headings, tables, lists, images, and links from the spatial layout. This is ideal for feeding documents to LLMs and RAG pipelines. This mode is purely heuristics and rule-based, so complex documents may not render perfectly, but it will be fast.

# Render to Markdown
lit parse document.pdf --format markdown -o output.md
Strip images instead of emitting placeholders
lit parse document.pdf --format markdown --image-mode off
Extract embedded images to disk and reference them from the markdown
lit parse document.pdf --format markdown --image-mode embed --extract-images --image-output-dir ./images
Extract image bytes and metadata without changing Markdown image handling
lit parse document.pdf --format json --extract-images
Emit link text as plain text (no text syntax)
lit parse document.pdf --format markdown --no-links
Include tagged-PDF logical structure in JSON
lit parse document.pdf --format json --extract-structure-tree
Include the classified layout blocks (with bounding boxes) in JSON
lit parse document.pdf --format json --extract-blocks

Image handling is controlled by --image-mode:

| Mode | Behavior | |------|----------| | placeholder (default) | Emits references in reading order | | off | Strips images entirely | | embed | Emits the same image references as placeholder |

--extract-images is the only option that enables embedded-image extraction. --image-output-dir requires it and writes the extracted bytes to disk. JSON output contains each image's name, path, page bbox, intrinsic pixel dimensions, rotation, format, and duplicate relationship; pixel bytes are never embedded in JSON. Identical image resources reuse the same output file.

Library callers can opt in with extract_images: true (Rust), extractImages: true (Node/WASM), or extract_images=True (Python). It defaults to false. Markdown image mode controls presentation only; placeholder refs are still discovered without bytes.

Markdown reconstruction quality varies with document complexity. For the

hardest documents (dense tables, multi-column layouts, scans),

LlamaParse

remains the most accurate option.

Vector Graphics

Vector path output is opt-in because path-heavy PDFs can produce large payloads. Enable it with --extract-vector-graphics, Rust/Python extract_vector_graphics = true, or JavaScript/WASM extractVectorGraphics: true. Each page then includes vector_graphics (vectorGraphics in JavaScript) with:

• shapes: path bounding box, stroke/fill paint state and ARGB colors, and

whether the path contains a Bezier curve.

• lines: compatible horizontal/vertical segments merged using stroke width

and paint colors, with top-left 72-DPI viewport coordinates.

The representation follows LlamaParse PDFium path extraction; LiteParse calls the shape rectangle bbox rather than PDFium's coords, and uses width / height rather than w / h. The field is absent (or None/undefined) by default. Diagonal and curved segments are represented by their parent shape but are not emitted as lines.

Tagged PDF structure tree

Enable --extract-structure-tree (Rust/Python extract_structure_tree, JavaScript/WASM extractStructureTree) to add a page-scoped structure_tree. It preserves every root and recursively exposes element type, ID, actual/alternate text, title, typed scalar attributes, marked-content IDs, children, and referenced link annotations. The field is absent by default; enabled untagged pages contain roots: [].

Layout blocks

The Markdown renderer works by classifying each page into blocks — headings, paragraphs, list items, tables, code, rules, figures — and then rendering them. Enable --extract-blocks (Rust/Python extract_blocks, JavaScript/WASM extractBlocks) to get that decomposition as data instead of only as rendered text, with the coordinates the classifier used.

Each page gains a blocks array in reading order — the same order, and the same blocks, the Markdown output is built from. Every block carries:

• kind: one of heading, paragraph, list_item, code, table,

grid_fallback, rule, figure.

• bbox: the region it occupies, in the same top-left 72-DPI viewport space as

text_items. This is the union of every source line that fed the block, so a wrapped heading or a multi-line paragraph reports its whole band.

• Kind-specific fields, omitted when they don't apply: text and level for

headings; ordered / marker for list items; lines and lang for code; header and rows for tables; id / format for figures.

Table cells are objects, not bare strings — each has text and its own bbox, so a cell can be mapped back to the region of the page it was read from. For ruled tables that box is the drawn grid cell; for borderless tables it is the extent of the spans the cell was built from. Cells that exist only to square off a ragged grid carry no bbox, since they have no ink behind them.

{
  "kind": "table",
  "bbox": { "x": 72.0, "y": 310.5, "width": 468.0, "height": 96.0 },
  "header": [
    { "text": "Territory Code", "bbox": { "x": 72.0, "y": 310.5, "width": 120.0, "height": 24.0 } },
    { "text": "Factor",         "bbox": { "x": 192.0, "y": 310.5, "width": 96.0, "height": 24.0 } }
  ],
  "rows": [
    [
      { "text": "001", "bbox": { "x": 72.0, "y": 334.5, "width": 120.0, "height": 24.0 } },
      { "text": "1.25", "bbox": { "x": 192.0, "y": 334.5, "width": 96.0, "height": 24.0 } }
    ]
  ]
}

Document metadata, content bounds, and XFA packets

Parse results (Rust/Node/Python APIs) carry the document's /Info creator and producer entries when present; these are API-only and never appear in CLI JSON output. Enable extract_document_metadata (JavaScript/WASM extractDocumentMetadata) to add doc_meta/docMeta, a provenance object with the /Info creation/modification dates, PDF version and encryption permissions, signature state, incremental-save markers, trailer ID comparison, the document catalog's XMP packet (capped at 64 KiB, with xmp_truncated when it was cut), and source file size. It is off by default because it streams the whole source file once; it is absent for inputs converted from a non-PDF format, where the facts would describe the intermediate PDF rather than your file. xmp needs a structural parse of the document, so it is skipped (left absent) for sources over 16 MiB and in WASM builds — the other fields are unaffected. Enable --extract-content-bounds (Rust/Python extract_content_bounds, JavaScript/WASM extractContentBounds) to add a per-page content_bounds: the union bbox of the page's top-level content objects in viewport coords (absent for empty pages). Enable --extract-xfa-packets (Rust/Python extract_xfa_packets, JavaScript/WASM extractXfaPackets) to add xfa_packets with each raw XFA packet's index, name, byte length, and XML content; non-XFA documents yield an empty list. All of these are off by default, so default JSON output is unchanged.

Screenshot raster signals

Screenshots draw AcroForm field appearances (filled values, checkbox states) on top of the page raster, so form data is visible in the render and to OCR. Each screenshot result reports is_solid_fill (blank page after render), and with detect_screenshot_rects (Node detectScreenshotRects) also rects: solid same-color rectangles and lines found in the raster in viewport coords, which covers scanned/flattened pages that carry no vector paths.

Check Complexity

Before committing to a full parse, check whether a document actually needs OCR or heavier processing. This is a cheap, text-layer-only pass — useful for routing documents to different pipelines, rejecting ones you can't handle, or estimating cost.

# Print the complexity verdict and per-page JSON
lit is-complex document.pdf
Use as a shell predicate — only parse with --no-ocr when the document is simple
lit is-complex document.pdf --quiet && lit parse document.pdf --no-ocr
List the pages that need OCR
lit is-complex document.pdf --compact | jq '[.[] | select(.needs_ocr) | .page_number]'

It always prints per-page JSON to stdout, a human-readable verdict to stderr, and exits non-zero when any page needs OCR. Each page carries a needs_ocr verdict and a list of reasons (scanned, no-text, sparse-text, embedded-images, garbled, vector-text, annotation-text).

Batch Parsing

Parse an entire directory of documents:

lit batch-parse ./input-directory ./output-directory

Generate Screenshots

Screenshots are essential for LLM agents to extract visual information that text alone cannot capture.

# Screenshot all pages
lit screenshot document.pdf -o ./screenshots
Screenshot specific pages
lit screenshot document.pdf --target-pages "1,3,5" -o ./screenshots
Custom DPI
lit screenshot document.pdf --dpi 300 -o ./screenshots

CLI Reference

Parse Command

``` lit parse [OPTIONS]

Options: -o, --output Output file path --format Output format: json|text|markdown [default: text] --no-ocr Disable OCR --ocr-language OCR language, Tesseract format [default: eng] --ocr-server-url HTTP OCR server URL (uses Tesseract if not