A small web app for testing OCR on images and PDFs.
The default setup uses Ollama with glm-ocr.
- Upload an image or PDF
- For PDFs, render the current page as an image before OCR
- Send the image to an OCR model
- Stream the result back into the page
- Render the output as markdown
-
Install Ollama.
-
Pull the OCR model:
ollama pull glm-ocr- Start Ollama.
On macOS and Windows, opening the Ollama app is usually enough.
On Linux, or if you are running it manually:
ollama serve- Install dependencies:
npm install- Copy the env file:
cp .env.example .env- Start the app:
npm run dev- Open the local URL from Vite and upload a file.
You can set the default OCR connection in .env.
VITE_OCR_BASE_URL=/api/proxy
VITE_OCR_ENDPOINT=/api/generate
VITE_OCR_MODEL=glm-ocr
OCR_PROXY_TARGET=http://127.0.0.1:11434What these mean:
VITE_OCR_BASE_URL: internal proxy base URL used by the browserVITE_OCR_ENDPOINT: OCR endpoint path, or a full URLVITE_OCR_MODEL: model name to send in the requestOCR_PROXY_TARGET: local dev proxy target for Vite
By default, the app calls /api/proxy/api/generate, and Vite proxies that to http://127.0.0.1:11434/api/generate.
This project is built around Ollama.
By default it uses Ollama's native /api/generate API and the glm-ocr model.
If you want to use a different setup, you can change the endpoint and model in .env or in the settings modal, but the easiest supported path is Ollama.
There is a settings button in the top-right corner.
It reads the default values from .env, then lets each user override them in the browser:
- OCR endpoint
- Model
These overrides are saved in localStorage, so they only affect that browser.
The UI does not write back into .env. That is expected. .env is a file used by the local dev/build environment, not something a browser app should edit directly.
npm run devnpm run buildnpm run lintnpm run check
MIT. See LICENSE.
