LiteParse: Document Parsing Without the LLM Tax

Attended a demo by Logan Markewich who walked through LiteParse a genuinely clever piece of engineering worth breaking down.
What makes it different: there's no LLM involved at all. Just clean markdown or JSON across PDF, DOCX, PPTX, XLSX and scanned images, at roughly 2ms per page.
For anyone building agent pipelines, that's the real value. Reading a document no longer requires a model call, which strips latency, cost, and a whole category of hallucination risk out of the ingestion step. Fast pipelines stay fast, and the agent gets usable structure immediately.
RAG ingestion is the obvious application. The more compelling one, in my view, is real-time agent tool calls where speed and determinism are exactly what the workflow demands.

Good post overall — the core claims check out, with a couple of nuances worth flagging.

Confirmed accurate:

• No LLM involved — correct. LiteParse's own docs describe it as running entirely on your machine with no cloud dependencies, no LLMs, and no API keys.
• Formats: PDF, DOCX, PPTX, XLSX, scanned images — confirmed. It parses PDFs, Office documents, and images, with built-in OCR support, and natively covers PDF, DOCX, PPTX, XLSX, or image via format auto-detection.
• Markdown or JSON output — confirmed, it can render to Markdown with headings, tables, lists, images, and links, or output structured JSON with bounding boxes.
• ~2ms/page speed — roughly checks out. LlamaIndex's own v2.0 benchmark reported just 0.777s for a 457 page 100MB document, which works out to about 1.7ms/page — in the ballpark of "roughly 2ms," though that's one large-document benchmark, not a guaranteed per-page figure across all doc types (OCR pages will run much slower than native-text pages).

Worth flagging:

• LiteParse isn't a from-scratch parser for DOCX/PPTX/XLSX — it converts them to PDF first (via LibreOffice) or, for images, via ImageMagick, then runs the same PDF pipeline: Office docs (DOCX, XLSX, PPTX): Converted to PDF via LibreOffice, then parsed with the same spatial pipeline. That's still LLM-free and fast, but it does mean there's a real system dependency on LibreOffice/ImageMagick being installed, not a pure in-process parse.
• It's explicitly positioned as the fast, rough-cut tier, not a full accuracy replacement for hard documents. LlamaIndex's own materials say for complex documents (dense tables, multi-column layouts, charts, handwritten text, or scanned PDFs), you'll get significantly better results with LlamaParse, their LLM-based cloud parser. So the "hallucination-free" framing is fair, but there's an implicit accuracy/speed tradeoff on messy documents that the post doesn't mention.
• It's built by LlamaIndex (Logan Markewich is their Head of Open Source), not an independent project — worth noting for context since he's presenting his own team's tool.

Your two applications (RAG ingestion vs. real-time agent tool calls) are both explicitly called out as intended use cases by the maintainers themselves, so that framing is on solid ground.