The short version
A long-document OCR approach using reference-sliding-window attention.
From the HackerLinks archive
Unlimited OCR
A long-document OCR approach using reference-sliding-window attention.
At a glance:
First seen:2026-06-23
Last seen:2026-06-23
Times seen:1
Website:github.com
Why it caught our attention
It sparked detailed discussion of long-document parsing, memory growth, and local AI deployments.
Where it surfaced on Hacker News
2026-06-23
Editorial paraphrase
Commenters liked the architecture for separating global reference from local generation and compared it to chunking and compaction.
Original threadUnlimited OCR: One-shot long-horizon parsing