Explanation

These pages exist to give you a mental model. They are not tutorials and they are not recipes — there are no commands to run, and nothing here is load-bearing for getting a working OCR pipeline up. Read them when you want to understand why the SDK is shaped the way it is.

Topics

• Layout and reading order — what a "block" is, how reading_order differs from layout, and why the Markdown renderer leans on blocks rather than raw OCR items.
• Searchable PDF internals — the invisible-text-overlay technique, why it preserves the original page, and where the font requirement comes from.
• HTTP vs gRPC — when each transport is the right pick, and the proto3 bool-presence caveat that subtly affects gRPC defaults.

Where to go next

• For step-by-step walkthroughs, see the tutorials.
• For short answers to specific problems, see the how-to guides.
• For exhaustive signatures, see the API reference.