Omnitome - Read 500-page books faster—with clickable source proof.
by•
Omnitome is a local-first AI reading app that compresses dense PDFs/EPUBs ~7:1 without losing theorems or formulas. Every sentence carries a clickable citation back to the verbatim source. It runs 100% offline on a 4GB laptop GPU, processing a 500,000-word book in just 5.9 minutes with zero data egress. Built for researchers and developers who refuse to trust hallucinated AI summaries.

Replies
Hey Product Hunt 👋
I built Omnitome because I was frustrated with the two bad options for reading massive texts.
On one side, you spend 20 hours reading a 500-page textbook or a 4,000-chapter web serial. On the other, you trust an AI summarizer that flattens the nuance, deletes the math, and hallucinates the plot.
Omnitome takes a third path: Compression, not summarization.
It’s a local-first reading app built on the "Two-Representation Invariant":
1. The Original Source is paper (immutable, authoritative).
2. The Omni Lens is tinted (densely compressed).
3. Derivation never masquerades as source (every sentence has a clickable citation back to the original text).
The hardest part was making this run 100% offline on my 4GB laptop GPU. I had to build a sequential await barrier that caps RAM at 296 MB, and a lifecycle manager that evicts idle PyTorch models to reclaim 3.2 GB of VRAM. A 508,000-word book now processes in 5.9 minutes, entirely offline.
It's open source, Windows-only for now, and completely free.
I'd love feedback from anyone who reads dense material for a living: What would you want a compression tool to preserve that standard AI summaries usually destroy?
— Ayush
https://github.com/ayumu-ml/Omnitome