LlamaIndex

LlamaIndex introduced ExtractBench as an open document-extraction benchmark, describing dataset scope, methodology, and the goal of evaluating extraction quality for enterprise document processing and agentic workflows.

LlamaIndex announced markdown support/tiering for LiteParse, including a note on benchmark-driven scope and instructions to install LiteParse across multiple runtimes (Node, Python, Rust, WASM).

LlamaIndex announced an expansion of LlamaParse Index with a Retrieval Harness designed to provide filesystem primitives, visual layout preservation, managed indexes, and pipeline observability for enterprise document agent navigation.

LlamaIndex announced granular bounding boxes in LlamaParse to improve attribution precision for extracted values in dense enterprise documents.

LlamaIndex described creating a deal sourcing agent using LlamaAgents Builder, using LlamaParse products for extraction/classification inputs into document-related workflows.

Jerry Liu discussed the shift toward long-horizon document agents, framing the opportunity for document-based knowledge work automation that can iterate over time and respond to changing context.

LlamaIndex discussed how “files” function as a core interface/abstraction for agentic systems—supporting long-running context, retrieval of external context at scale, and providing a way to equip agents with skills.

LlamaIndex announced a $19 million Series A led by Norwest Venture Partners with participation from Greylock, alongside the launch and general availability of LlamaCloud.