Android / PDF・Image・OCR / Offline-first

ClarityPDF

ClarityPDF is an Android app that brings PDF creation and editing, conversion and composition, and OCR / AI-OCR together in one place. Processing completes on the device as a rule; input files are read-only and results are always saved as new files.

Documents: User manual | Privacy policy

        Input is read-only and non-destructive. Every edit or conversion is written out as a new file.
OCR distinguishes "embedded text" (already in the file) from "OCR" (image recognition), and can export JSON.
The LLM connection defaults to on-device 127.0.0.1. Settings include a "Launch LLM API" button and a status tile.

      

Screenshots

Home screen: the ClarityPDF title and four categories — Create & edit PDF, Convert & compose PDF, OCR / AI-OCR, and Settings. — Home screen. Four categories — Create & edit PDF, Convert & compose PDF, OCR / AI-OCR, and Settings — lead to each tool.

Edit PDF screen: a fixed canvas with an added image and styled text, and a bottom toolbox with Text / Image / Layers / Font / Save. — Edit PDF screen. The canvas stays fixed while you add and edit from the bottom toolbox (Text / Image / Layers / Font / Save) and floating windows. Shown with an image and underlined text placed.

AI Prompt screen: instruction field, Processing unit (Whole PDF / Per page), method (OCR→LLM / Vision), and OCR engine selection. — AI Prompt. Enter an instruction and choose the processing unit (Whole PDF / Per page), the method (OCR→LLM / Vision), and the OCR engine.

AI Prompt screen (continued): LLM connection info (127.0.0.1:11434 / Qwen), the Run AI button, and the whole-document result. — The lower part of the same screen. Confirm the LLM endpoint (on-device 127.0.0.1 by default) and run to see the result for the whole PDF.

Key features

Edit PDF: add text/images and replace or delete existing text. The canvas stays fixed while you add and edit from floating windows and a toolbox. Bold / italic / underline, rotation, scaling, links (URLs), and font selection (several OFL fonts, per text run) are supported.
Split (extract pages): combine selected pages into one PDF, or split into one page each.
Merge: combine several PDFs into one in a chosen order.
Reorder: drag thumbnails to change the page order.
PDF to images: convert each page to PNG / JPEG (DPI configurable).
Images to PDF: combine several images into one PDF in a chosen order.

OCR / AI-OCR

OCR / text extraction: extract text from PDFs/images. Embedded text and OCR are distinguished and can be exported as JSON. Engines: Tesseract / PaddleOCR / local LLM Vision, with Japanese, English, Simplified Chinese, and Korean recognition.
PDF summary: summarize the whole file or per page with an LLM. Choose "OCR→LLM" or "Vision (page images sent directly to the LLM)".
AI Prompt: enter an arbitrary instruction (prompt) and run it over the whole PDF or per page (OCR→LLM / Vision).
LLM connection: works with Ollama / OpenAI-compatible servers. Fetch the model list via /api/tags, test the connection, launch the API, and see its status.

Notes

Adding/editing text needs an embedded font. Choose from Noto Sans JP / Noto Serif JP / M PLUS Rounded 1c / Zen Kaku Gothic New / Klee One (all SIL OFL), per text run — each downloads once, then used offline.
The only network use is "the first download of OCR models / fonts" and "sending to the configured server when using an LLM". The default LLM endpoint is on-device 127.0.0.1.
If you configure an external LLM server, page images and extracted text are sent there and handled under that server's policies.