vLLM Studio launches local OCR desktop app

This is the missing UX layer for the local OCR wave: model releases are moving fast, but the tooling to actually try them on messy documents has lagged behind. By wrapping ingestion, layout parsing, and markdown cleanup into one app, vLLM Studio makes local multimodal testing feel like a product instead of a script. The local-first architecture is the real win here because sensitive documents stay on your machine, which matters for enterprise, legal, and research workflows. Layout-aware extraction is more important than raw OCR in practice; preserving tables, captions, math, and code is what makes downstream LLM use reliable. First-party presets for Chandra, GLM OCR, and LightOn reduce the usual model-compatibility roulette and make the app more immediately useful. The tradeoff is setup friction: it still assumes you have a local vLLM backend ready, and Linux/Windows support is only listed as coming soon.