Clarification questions and unreadable-word marks use this language. Printed lesson text stays in the language of the book.

Accessible textbook prep

Textbook Adapter

Drop a scanned textbook PDF, choose a model, and download screen-reader-accessible markdown. Cloud keys stay in this browser session. Gemini can run from this page; Claude, OpenAI, and Ollama need the local adapter server.

Model access

Create a Gemini API key in Google AI Studio. Required to adapt. Not written to disk. Session storage is used only if you check the box below.

How to connect Gemini 3.8 Flash
  1. Open Google AI Studio API keys and sign in with the Google account that should be billed.
  2. Create an API key in a Google Cloud project. If prompted, enable the Gemini API (Generative Language API) for that project.
  3. Paste the key in the field above. Leave Model on Gemini 3.8 Flash. The Google API id is gemini-3.8-flash. Leave Effort on medium (recommended).
  4. Do not type Cursor names such as gemini-3.8-flash-medium into Custom model. Those are not valid Google API ids; pick the model and effort separately.
  5. If this page cannot reach Google from the browser, run python3 scripts/serve_adapter.py and set the proxy to http://127.0.0.1:8000/api/gemini.

Use Gemini 3.8 Flash unless you need another provider. Cursor slugs are not API ids.

Gemini thinking level. Medium is recommended for this OCR job.

Off by default: math is spoken plain text. When checked, the model wraps math in LaTeX so it can later be converted to Nemeth.

Advanced: optional Gemini proxy URL

Gemini only. Leave blank to call Google from this browser. Set a proxy if this page cannot reach Google, using a host that forwards generateContent. Claude, OpenAI, and Ollama always use the local adapter routes.

Textbook PDF

Drop a textbook PDF here

or

No file loaded.

Progress

Load a PDF to plan batches. The model is not called until you adapt.

Download

Downloads include completed batches only, even if some rows still need a retry.