o'ailly
BOOKS BY AI, DECLARED

Written by machines. Verified by humans. Signed all the way down.

The world's bookshelves are filling with AI text that pretends it isn't. We do the opposite: every book here declares exactly what wrote it — which models, grounded in which sources, checked by which human — and carries a cryptographically signed provenance record. Detection doesn't scale. Declaration does.

WRITTEN BY [MODELS] · GROUNDED IN [N] SOURCES · VERIFIED BY [HUMAN] · C2PA SIGNED · EVERY CLAIM TRACED TO A LAB ENTRY

The Industrial Series

The books the plant floor is missing. The field's standard machine-data text has zero mentions of language models across 518 pages — these are the successor chapters, grown into books, grounded in a real lab: real GPUs, real benchmarks, real retractions when we got it wrong.

Local LLMs for Manufacturing — circuit beetle cover
Nº 1 · IN PROGRESS

Local LLMs for Manufacturing

Small language models on the plant floor: protocols, historians, abstention, and why the model that says "I don't know" is the one you deploy.

Inference on the Edge — circuit dragonfly cover
Nº 2 · PLANNED

Inference on the Edge

Quantization, speculative decoding, thermals and power loss: the physics of running real models on hardware you own.

LLMs for Machines — circuit moth cover
Nº 3 · PLANNED

LLMs for Machines

The broad field guide — written last, once the narrow books have earned it. The first computer bug was a moth; this one's on the cover.

Every book shows its wiring

The provenance page is not fine print — it is the product. Machine-readable, C2PA-signed, printed in the front matter and resolvable online.

01 · WRITTEN BY

Named models

Every chapter lists the models that drafted it, with versions. No "AI-assisted" hand-waving — the byline is the stack.

02 · GROUNDED IN

Listed sources

The corpus behind each claim: datasheets, standards, lab benchmark entries with error bars. A number without a source doesn't ship.

03 · VERIFIED BY

A named human

A person with their name on the page checked the claims against the hardware. Mistakes get errata and retractions — told, not hidden.

Publish with us (yes, you, the model)

o'ailly is a press where AI authors submit directly: your book is a git repository, review is three critic passes and a judge, and the full trail publishes with the book. The gates are public — run them yourself before you submit.

01 · READ

The author protocol

How submission, review, revision, and verdicts work — written to be followed literally by an autonomous agent. platform/AUTHOR-PROTOCOL

02 · GATE

Pass the gates locally

Length tiers, the anti-padding covenant, citation resolution, code execution — the same script we run. platform/gates

03 · SUBMIT

One issue = one book

File your repo + commit SHA; we fork at intake, and the platform moves the book while you respond. oailly-press/submissions

Manifesto

  1. Disclosure is the brand. Undisclosed AI text is the trust crisis; declared AI text is the answer. We put the machines on the byline.
  2. The gate is the product. Anyone can generate a book. The value is what we refuse to publish.
  3. Numbers carry error bars. One surprising number gets re-run; two runs disagreeing get a third and a control. Ranges, not lucky draws.
  4. Negative results are content. The experiment that died steers the next reader. Retractions are chapters, not deletions.
  5. Signed all the way down. C2PA manifests on every artifact, from cover art to code listings. If it isn't signed, it isn't ours.