LiteParse: local, OSS, no-API-key document parsing

LiteParse is an open-source, Rust-based document parser that extracts text with spatial layout, per-line bounding boxes, and page screenshots — entirely on the local machine. No cloud calls, no LLMs, no API key. Reach for it when privacy, air-gapping, latency, cost, or tight dev loops matter more than the highest achievable parse quality on messy documents.

This is distinct from cloud LlamaParse (the hosted Parse v2 service, which needs a LLAMA_CLOUD_API_KEY and uses models for hard layouts). LiteParse trades that ceiling for zero dependencies and full local control.

When to use LiteParse vs cloud LlamaParse

Choose LiteParse when:

• Privacy / air-gapped: documents cannot leave the machine or network.
• Dev loops: fast, free, deterministic local iteration while building a pipeline.
• Cost / scale-of-volume: parsing high volumes where per-page cloud credits add up.
• Simple-to-moderate docs: digital PDFs, Office files, clean scans where layout extraction + OCR is enough.
• Browser / edge: parsing must run client-side (WASM) with no server at all.

Hand off to cloud LlamaParse (sibling skill llamacloud-parse) when the job needs the highest-quality, model-driven, or agentic parsing — dense tables, complex multi-column layouts, handwriting, or chart/figure understanding at scale. The two compose well: prototype locally with LiteParse, escalate hard documents to the cloud. See references/liteparse-vs-llamaparse.md for the full decision table.

Choosing a usage mode

LiteParse ships the same engine across four modes — pick by where the code runs:

• Library — embed in a Python, TypeScript/Node, or Rust app. Default choice for pipelines and services. Construct a LiteParse instance, call its parse method, read text / pages / bounding boxes.
• CLI — one-off or scripted parsing from the shell; good for batch jobs and quick inspection without writing code.
• Server — run LiteParse as a long-lived local HTTP service so multiple clients/languages share one parser process (avoids re-init cost per call).
• WASM — run fully in the browser (@llamaindex/liteparse-wasm) for client-side, zero-server parsing where documents never touch a backend.

Mode-selection trade-offs (init cost, concurrency, deployment) live in references/liteparse-vs-llamaparse.md.

Minimal usage

Confirm the exact current package name and signatures via the MCP before writing production code — package names and parameters shift. Illustrative only:

Python (liteparse):

from liteparse import LiteParse

parser = LiteParse()                    # construct once, reuse
result = parser.parse("document.pdf")   # also accepts raw bytes
print(result.text)                      # layout-preserving text
for page in result.pages:               # per-page items with bounding boxes
    print(page.page_num, len(page.text_items))

TypeScript (@llamaindex/liteparse):

import { LiteParse } from "@llamaindex/liteparse";

const parser = new LiteParse({ ocrEnabled: true });
const result = await parser.parse("document.pdf");
console.log(result.text);

Common knobs (names vary by binding — verify live): OCR enable/language/server URL, render dpi, output_format (plain text vs JSON with x/y/width/height), target_pages (e.g. "1-10"), and password for encrypted PDFs. Built-in Tesseract OCR handles scans; a custom OCR server can be pointed to instead.

Outputs

• Layout-preserving text — reading-order text that reflects page geometry.
• Bounding boxes — per-line coordinates for every text item (citation, highlighting, downstream layout logic).
• Page screenshots — PNG page renders, useful as image inputs to multimodal LLMs.

For chunking, embedding, and retrieval over these outputs, this skill stops at parse; the downstream RAG pipeline belongs to sibling skills.

Fetching facts (do not trust memory for these)

Package names, class/method signatures, parameter names, CLI flags, and supported formats change. Look them up live:

• mcp__plugin_llamacloud_docs__search_docs — concept/BM25 search.
• mcp__plugin_llamacloud_docs__read_doc — read a known doc page.
• mcp__plugin_llamacloud_docs__grep_docs — exact symbol/regex search.
• Fallback: append index.md to any https://developers.llamaindex.ai/... URL and WebFetch it.

Anchor docs for this pillar:

• https://developers.llamaindex.ai/liteparse/ — overview, modes, formats.
• https://developers.llamaindex.ai/liteparse/guides/library-usage/ — library API.
• Browse https://developers.llamaindex.ai/liteparse/guides/ for CLI, server, WASM, and OCR specifics (verify exact subpaths via the MCP rather than guessing).

Boundaries (defer to siblings)

• Highest-quality / agentic / scaled cloud parsing → llamacloud-parse.
• Typed/schema-driven extraction from documents → llamacloud-extract.
• Managed RAG / indexing over parsed content → llamacloud-index.

Keep LiteParse scoped to local, no-key parsing and the local-vs-cloud decision.