LlamaIndex open-sources LiteParse for local document parsing

This feels like the right abstraction for a lot of agent workflows: not perfect document understanding, but fast, local, and “good enough” output that an LLM can actually use immediately.

• –The big win is latency and portability: agents can parse PDFs, Office docs, and images locally instead of spawning ad hoc Python parsing code or waiting on hosted APIs.
• –Preserving spatial layout instead of aggressively reconstructing structure is a smart bet for LLMs, especially for tables, indentation, and other ASCII-friendly formats.
• –Screenshot support makes it more than a text extractor; it gives agents a fallback path when visual reasoning matters.
• –The built-in OCR story is pragmatic: Tesseract.js by default, with optional PaddleOCR or EasyOCR servers for harder scans.
• –LlamaIndex is also drawing a clear product line: LiteParse handles common, fast, agentic parsing, while LlamaParse remains the better choice for messy, high-stakes documents.