Local LLM PDF OCR
A tool that turns scanned PDFs and images into searchable documents by reading them with a locally run vision language model instead of traditional OCR.
Local LLM PDF OCR converts scanned PDFs and images into searchable, selectable documents using a local vision language model such as olmOCR, Qwen3-VL, or GLM-OCR, rather than a traditional engine like Tesseract. It writes an invisible, selectable text layer back under the original page image, and is driven from a configurable web UI or a command-line interface.
A layout detector finds text boxes on the page, the vision model transcribes the whole page, and a dynamic-programming aligner binds the model's output lines to the correct boxes in reading order, with a per-box re-OCR fallback for boxes it cannot confidently place. An opt-in grounded path instead uses a model that returns text with coordinates directly, and a text-only fast path skips layout detection and alignment entirely for raw text output.
It accepts PDF, JPEG, PNG, BMP, WEBP, TIFF, and AVIF input, runs fully offline through LM Studio, Ollama, vLLM, or any OpenAI-compatible endpoint, and outputs a searchable PDF, HTML, Markdown, or plain text.